2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
44 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
45 struct btrfs_root
*root
,
46 struct btrfs_device
*device
);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
48 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
49 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
51 static DEFINE_MUTEX(uuid_mutex
);
52 static LIST_HEAD(fs_uuids
);
54 static void lock_chunks(struct btrfs_root
*root
)
56 mutex_lock(&root
->fs_info
->chunk_mutex
);
59 static void unlock_chunks(struct btrfs_root
*root
)
61 mutex_unlock(&root
->fs_info
->chunk_mutex
);
64 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
66 struct btrfs_device
*device
;
67 WARN_ON(fs_devices
->opened
);
68 while (!list_empty(&fs_devices
->devices
)) {
69 device
= list_entry(fs_devices
->devices
.next
,
70 struct btrfs_device
, dev_list
);
71 list_del(&device
->dev_list
);
72 rcu_string_free(device
->name
);
78 static void btrfs_kobject_uevent(struct block_device
*bdev
,
79 enum kobject_action action
)
83 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
85 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
87 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
88 &disk_to_dev(bdev
->bd_disk
)->kobj
);
91 void btrfs_cleanup_fs_uuids(void)
93 struct btrfs_fs_devices
*fs_devices
;
95 while (!list_empty(&fs_uuids
)) {
96 fs_devices
= list_entry(fs_uuids
.next
,
97 struct btrfs_fs_devices
, list
);
98 list_del(&fs_devices
->list
);
99 free_fs_devices(fs_devices
);
103 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
106 struct btrfs_device
*dev
;
108 list_for_each_entry(dev
, head
, dev_list
) {
109 if (dev
->devid
== devid
&&
110 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
117 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
119 struct btrfs_fs_devices
*fs_devices
;
121 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
122 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
129 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
130 int flush
, struct block_device
**bdev
,
131 struct buffer_head
**bh
)
135 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
138 ret
= PTR_ERR(*bdev
);
139 printk(KERN_INFO
"btrfs: open %s failed\n", device_path
);
144 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
145 ret
= set_blocksize(*bdev
, 4096);
147 blkdev_put(*bdev
, flags
);
150 invalidate_bdev(*bdev
);
151 *bh
= btrfs_read_dev_super(*bdev
);
154 blkdev_put(*bdev
, flags
);
166 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
167 struct bio
*head
, struct bio
*tail
)
170 struct bio
*old_head
;
172 old_head
= pending_bios
->head
;
173 pending_bios
->head
= head
;
174 if (pending_bios
->tail
)
175 tail
->bi_next
= old_head
;
177 pending_bios
->tail
= tail
;
181 * we try to collect pending bios for a device so we don't get a large
182 * number of procs sending bios down to the same device. This greatly
183 * improves the schedulers ability to collect and merge the bios.
185 * But, it also turns into a long list of bios to process and that is sure
186 * to eventually make the worker thread block. The solution here is to
187 * make some progress and then put this work struct back at the end of
188 * the list if the block device is congested. This way, multiple devices
189 * can make progress from a single worker thread.
191 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
194 struct backing_dev_info
*bdi
;
195 struct btrfs_fs_info
*fs_info
;
196 struct btrfs_pending_bios
*pending_bios
;
200 unsigned long num_run
;
201 unsigned long batch_run
= 0;
203 unsigned long last_waited
= 0;
205 int sync_pending
= 0;
206 struct blk_plug plug
;
209 * this function runs all the bios we've collected for
210 * a particular device. We don't want to wander off to
211 * another device without first sending all of these down.
212 * So, setup a plug here and finish it off before we return
214 blk_start_plug(&plug
);
216 bdi
= blk_get_backing_dev_info(device
->bdev
);
217 fs_info
= device
->dev_root
->fs_info
;
218 limit
= btrfs_async_submit_limit(fs_info
);
219 limit
= limit
* 2 / 3;
222 spin_lock(&device
->io_lock
);
227 /* take all the bios off the list at once and process them
228 * later on (without the lock held). But, remember the
229 * tail and other pointers so the bios can be properly reinserted
230 * into the list if we hit congestion
232 if (!force_reg
&& device
->pending_sync_bios
.head
) {
233 pending_bios
= &device
->pending_sync_bios
;
236 pending_bios
= &device
->pending_bios
;
240 pending
= pending_bios
->head
;
241 tail
= pending_bios
->tail
;
242 WARN_ON(pending
&& !tail
);
245 * if pending was null this time around, no bios need processing
246 * at all and we can stop. Otherwise it'll loop back up again
247 * and do an additional check so no bios are missed.
249 * device->running_pending is used to synchronize with the
252 if (device
->pending_sync_bios
.head
== NULL
&&
253 device
->pending_bios
.head
== NULL
) {
255 device
->running_pending
= 0;
258 device
->running_pending
= 1;
261 pending_bios
->head
= NULL
;
262 pending_bios
->tail
= NULL
;
264 spin_unlock(&device
->io_lock
);
269 /* we want to work on both lists, but do more bios on the
270 * sync list than the regular list
273 pending_bios
!= &device
->pending_sync_bios
&&
274 device
->pending_sync_bios
.head
) ||
275 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
276 device
->pending_bios
.head
)) {
277 spin_lock(&device
->io_lock
);
278 requeue_list(pending_bios
, pending
, tail
);
283 pending
= pending
->bi_next
;
286 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
287 waitqueue_active(&fs_info
->async_submit_wait
))
288 wake_up(&fs_info
->async_submit_wait
);
290 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
293 * if we're doing the sync list, record that our
294 * plug has some sync requests on it
296 * If we're doing the regular list and there are
297 * sync requests sitting around, unplug before
300 if (pending_bios
== &device
->pending_sync_bios
) {
302 } else if (sync_pending
) {
303 blk_finish_plug(&plug
);
304 blk_start_plug(&plug
);
308 btrfsic_submit_bio(cur
->bi_rw
, cur
);
315 * we made progress, there is more work to do and the bdi
316 * is now congested. Back off and let other work structs
319 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
320 fs_info
->fs_devices
->open_devices
> 1) {
321 struct io_context
*ioc
;
323 ioc
= current
->io_context
;
326 * the main goal here is that we don't want to
327 * block if we're going to be able to submit
328 * more requests without blocking.
330 * This code does two great things, it pokes into
331 * the elevator code from a filesystem _and_
332 * it makes assumptions about how batching works.
334 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
335 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
337 ioc
->last_waited
== last_waited
)) {
339 * we want to go through our batch of
340 * requests and stop. So, we copy out
341 * the ioc->last_waited time and test
342 * against it before looping
344 last_waited
= ioc
->last_waited
;
349 spin_lock(&device
->io_lock
);
350 requeue_list(pending_bios
, pending
, tail
);
351 device
->running_pending
= 1;
353 spin_unlock(&device
->io_lock
);
354 btrfs_requeue_work(&device
->work
);
357 /* unplug every 64 requests just for good measure */
358 if (batch_run
% 64 == 0) {
359 blk_finish_plug(&plug
);
360 blk_start_plug(&plug
);
369 spin_lock(&device
->io_lock
);
370 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
372 spin_unlock(&device
->io_lock
);
375 blk_finish_plug(&plug
);
378 static void pending_bios_fn(struct btrfs_work
*work
)
380 struct btrfs_device
*device
;
382 device
= container_of(work
, struct btrfs_device
, work
);
383 run_scheduled_bios(device
);
386 static noinline
int device_list_add(const char *path
,
387 struct btrfs_super_block
*disk_super
,
388 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
390 struct btrfs_device
*device
;
391 struct btrfs_fs_devices
*fs_devices
;
392 struct rcu_string
*name
;
393 u64 found_transid
= btrfs_super_generation(disk_super
);
395 fs_devices
= find_fsid(disk_super
->fsid
);
397 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
400 INIT_LIST_HEAD(&fs_devices
->devices
);
401 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
402 list_add(&fs_devices
->list
, &fs_uuids
);
403 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
404 fs_devices
->latest_devid
= devid
;
405 fs_devices
->latest_trans
= found_transid
;
406 mutex_init(&fs_devices
->device_list_mutex
);
409 device
= __find_device(&fs_devices
->devices
, devid
,
410 disk_super
->dev_item
.uuid
);
413 if (fs_devices
->opened
)
416 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
418 /* we can safely leave the fs_devices entry around */
421 device
->devid
= devid
;
422 device
->dev_stats_valid
= 0;
423 device
->work
.func
= pending_bios_fn
;
424 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
426 spin_lock_init(&device
->io_lock
);
428 name
= rcu_string_strdup(path
, GFP_NOFS
);
433 rcu_assign_pointer(device
->name
, name
);
434 INIT_LIST_HEAD(&device
->dev_alloc_list
);
436 /* init readahead state */
437 spin_lock_init(&device
->reada_lock
);
438 device
->reada_curr_zone
= NULL
;
439 atomic_set(&device
->reada_in_flight
, 0);
440 device
->reada_next
= 0;
441 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
442 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
444 mutex_lock(&fs_devices
->device_list_mutex
);
445 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
446 mutex_unlock(&fs_devices
->device_list_mutex
);
448 device
->fs_devices
= fs_devices
;
449 fs_devices
->num_devices
++;
450 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
451 name
= rcu_string_strdup(path
, GFP_NOFS
);
454 rcu_string_free(device
->name
);
455 rcu_assign_pointer(device
->name
, name
);
456 if (device
->missing
) {
457 fs_devices
->missing_devices
--;
462 if (found_transid
> fs_devices
->latest_trans
) {
463 fs_devices
->latest_devid
= devid
;
464 fs_devices
->latest_trans
= found_transid
;
466 *fs_devices_ret
= fs_devices
;
470 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
472 struct btrfs_fs_devices
*fs_devices
;
473 struct btrfs_device
*device
;
474 struct btrfs_device
*orig_dev
;
476 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
478 return ERR_PTR(-ENOMEM
);
480 INIT_LIST_HEAD(&fs_devices
->devices
);
481 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
482 INIT_LIST_HEAD(&fs_devices
->list
);
483 mutex_init(&fs_devices
->device_list_mutex
);
484 fs_devices
->latest_devid
= orig
->latest_devid
;
485 fs_devices
->latest_trans
= orig
->latest_trans
;
486 fs_devices
->total_devices
= orig
->total_devices
;
487 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
489 /* We have held the volume lock, it is safe to get the devices. */
490 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
491 struct rcu_string
*name
;
493 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
498 * This is ok to do without rcu read locked because we hold the
499 * uuid mutex so nothing we touch in here is going to disappear.
501 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
506 rcu_assign_pointer(device
->name
, name
);
508 device
->devid
= orig_dev
->devid
;
509 device
->work
.func
= pending_bios_fn
;
510 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
511 spin_lock_init(&device
->io_lock
);
512 INIT_LIST_HEAD(&device
->dev_list
);
513 INIT_LIST_HEAD(&device
->dev_alloc_list
);
515 list_add(&device
->dev_list
, &fs_devices
->devices
);
516 device
->fs_devices
= fs_devices
;
517 fs_devices
->num_devices
++;
521 free_fs_devices(fs_devices
);
522 return ERR_PTR(-ENOMEM
);
525 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
526 struct btrfs_fs_devices
*fs_devices
, int step
)
528 struct btrfs_device
*device
, *next
;
530 struct block_device
*latest_bdev
= NULL
;
531 u64 latest_devid
= 0;
532 u64 latest_transid
= 0;
534 mutex_lock(&uuid_mutex
);
536 /* This is the initialized path, it is safe to release the devices. */
537 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
538 if (device
->in_fs_metadata
) {
539 if (!device
->is_tgtdev_for_dev_replace
&&
541 device
->generation
> latest_transid
)) {
542 latest_devid
= device
->devid
;
543 latest_transid
= device
->generation
;
544 latest_bdev
= device
->bdev
;
549 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
551 * In the first step, keep the device which has
552 * the correct fsid and the devid that is used
553 * for the dev_replace procedure.
554 * In the second step, the dev_replace state is
555 * read from the device tree and it is known
556 * whether the procedure is really active or
557 * not, which means whether this device is
558 * used or whether it should be removed.
560 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
565 blkdev_put(device
->bdev
, device
->mode
);
567 fs_devices
->open_devices
--;
569 if (device
->writeable
) {
570 list_del_init(&device
->dev_alloc_list
);
571 device
->writeable
= 0;
572 if (!device
->is_tgtdev_for_dev_replace
)
573 fs_devices
->rw_devices
--;
575 list_del_init(&device
->dev_list
);
576 fs_devices
->num_devices
--;
577 rcu_string_free(device
->name
);
581 if (fs_devices
->seed
) {
582 fs_devices
= fs_devices
->seed
;
586 fs_devices
->latest_bdev
= latest_bdev
;
587 fs_devices
->latest_devid
= latest_devid
;
588 fs_devices
->latest_trans
= latest_transid
;
590 mutex_unlock(&uuid_mutex
);
593 static void __free_device(struct work_struct
*work
)
595 struct btrfs_device
*device
;
597 device
= container_of(work
, struct btrfs_device
, rcu_work
);
600 blkdev_put(device
->bdev
, device
->mode
);
602 rcu_string_free(device
->name
);
606 static void free_device(struct rcu_head
*head
)
608 struct btrfs_device
*device
;
610 device
= container_of(head
, struct btrfs_device
, rcu
);
612 INIT_WORK(&device
->rcu_work
, __free_device
);
613 schedule_work(&device
->rcu_work
);
616 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
618 struct btrfs_device
*device
;
620 if (--fs_devices
->opened
> 0)
623 mutex_lock(&fs_devices
->device_list_mutex
);
624 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
625 struct btrfs_device
*new_device
;
626 struct rcu_string
*name
;
629 fs_devices
->open_devices
--;
631 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
632 list_del_init(&device
->dev_alloc_list
);
633 fs_devices
->rw_devices
--;
636 if (device
->can_discard
)
637 fs_devices
->num_can_discard
--;
639 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
640 BUG_ON(!new_device
); /* -ENOMEM */
641 memcpy(new_device
, device
, sizeof(*new_device
));
643 /* Safe because we are under uuid_mutex */
645 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
646 BUG_ON(device
->name
&& !name
); /* -ENOMEM */
647 rcu_assign_pointer(new_device
->name
, name
);
649 new_device
->bdev
= NULL
;
650 new_device
->writeable
= 0;
651 new_device
->in_fs_metadata
= 0;
652 new_device
->can_discard
= 0;
653 spin_lock_init(&new_device
->io_lock
);
654 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
656 call_rcu(&device
->rcu
, free_device
);
658 mutex_unlock(&fs_devices
->device_list_mutex
);
660 WARN_ON(fs_devices
->open_devices
);
661 WARN_ON(fs_devices
->rw_devices
);
662 fs_devices
->opened
= 0;
663 fs_devices
->seeding
= 0;
668 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
670 struct btrfs_fs_devices
*seed_devices
= NULL
;
673 mutex_lock(&uuid_mutex
);
674 ret
= __btrfs_close_devices(fs_devices
);
675 if (!fs_devices
->opened
) {
676 seed_devices
= fs_devices
->seed
;
677 fs_devices
->seed
= NULL
;
679 mutex_unlock(&uuid_mutex
);
681 while (seed_devices
) {
682 fs_devices
= seed_devices
;
683 seed_devices
= fs_devices
->seed
;
684 __btrfs_close_devices(fs_devices
);
685 free_fs_devices(fs_devices
);
688 * Wait for rcu kworkers under __btrfs_close_devices
689 * to finish all blkdev_puts so device is really
690 * free when umount is done.
696 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
697 fmode_t flags
, void *holder
)
699 struct request_queue
*q
;
700 struct block_device
*bdev
;
701 struct list_head
*head
= &fs_devices
->devices
;
702 struct btrfs_device
*device
;
703 struct block_device
*latest_bdev
= NULL
;
704 struct buffer_head
*bh
;
705 struct btrfs_super_block
*disk_super
;
706 u64 latest_devid
= 0;
707 u64 latest_transid
= 0;
714 list_for_each_entry(device
, head
, dev_list
) {
720 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
725 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
726 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
727 if (devid
!= device
->devid
)
730 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
734 device
->generation
= btrfs_super_generation(disk_super
);
735 if (!latest_transid
|| device
->generation
> latest_transid
) {
736 latest_devid
= devid
;
737 latest_transid
= device
->generation
;
741 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
742 device
->writeable
= 0;
744 device
->writeable
= !bdev_read_only(bdev
);
748 q
= bdev_get_queue(bdev
);
749 if (blk_queue_discard(q
)) {
750 device
->can_discard
= 1;
751 fs_devices
->num_can_discard
++;
755 device
->in_fs_metadata
= 0;
756 device
->mode
= flags
;
758 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
759 fs_devices
->rotating
= 1;
761 fs_devices
->open_devices
++;
762 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
763 fs_devices
->rw_devices
++;
764 list_add(&device
->dev_alloc_list
,
765 &fs_devices
->alloc_list
);
772 blkdev_put(bdev
, flags
);
775 if (fs_devices
->open_devices
== 0) {
779 fs_devices
->seeding
= seeding
;
780 fs_devices
->opened
= 1;
781 fs_devices
->latest_bdev
= latest_bdev
;
782 fs_devices
->latest_devid
= latest_devid
;
783 fs_devices
->latest_trans
= latest_transid
;
784 fs_devices
->total_rw_bytes
= 0;
789 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
790 fmode_t flags
, void *holder
)
794 mutex_lock(&uuid_mutex
);
795 if (fs_devices
->opened
) {
796 fs_devices
->opened
++;
799 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
801 mutex_unlock(&uuid_mutex
);
806 * Look for a btrfs signature on a device. This may be called out of the mount path
807 * and we are not allowed to call set_blocksize during the scan. The superblock
808 * is read via pagecache
810 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
811 struct btrfs_fs_devices
**fs_devices_ret
)
813 struct btrfs_super_block
*disk_super
;
814 struct block_device
*bdev
;
825 * we would like to check all the supers, but that would make
826 * a btrfs mount succeed after a mkfs from a different FS.
827 * So, we need to add a special mount option to scan for
828 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
830 bytenr
= btrfs_sb_offset(0);
832 mutex_lock(&uuid_mutex
);
834 bdev
= blkdev_get_by_path(path
, flags
, holder
);
841 /* make sure our super fits in the device */
842 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
845 /* make sure our super fits in the page */
846 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
849 /* make sure our super doesn't straddle pages on disk */
850 index
= bytenr
>> PAGE_CACHE_SHIFT
;
851 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
854 /* pull in the page with our super */
855 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
858 if (IS_ERR_OR_NULL(page
))
863 /* align our pointer to the offset of the super block */
864 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
866 if (btrfs_super_bytenr(disk_super
) != bytenr
||
867 disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
))
870 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
871 transid
= btrfs_super_generation(disk_super
);
872 total_devices
= btrfs_super_num_devices(disk_super
);
874 if (disk_super
->label
[0]) {
875 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
876 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
877 printk(KERN_INFO
"device label %s ", disk_super
->label
);
879 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
882 printk(KERN_CONT
"devid %llu transid %llu %s\n",
883 (unsigned long long)devid
, (unsigned long long)transid
, path
);
885 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
886 if (!ret
&& fs_devices_ret
)
887 (*fs_devices_ret
)->total_devices
= total_devices
;
891 page_cache_release(page
);
894 blkdev_put(bdev
, flags
);
896 mutex_unlock(&uuid_mutex
);
900 /* helper to account the used device space in the range */
901 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
902 u64 end
, u64
*length
)
904 struct btrfs_key key
;
905 struct btrfs_root
*root
= device
->dev_root
;
906 struct btrfs_dev_extent
*dev_extent
;
907 struct btrfs_path
*path
;
911 struct extent_buffer
*l
;
915 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
918 path
= btrfs_alloc_path();
923 key
.objectid
= device
->devid
;
925 key
.type
= BTRFS_DEV_EXTENT_KEY
;
927 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
931 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
938 slot
= path
->slots
[0];
939 if (slot
>= btrfs_header_nritems(l
)) {
940 ret
= btrfs_next_leaf(root
, path
);
948 btrfs_item_key_to_cpu(l
, &key
, slot
);
950 if (key
.objectid
< device
->devid
)
953 if (key
.objectid
> device
->devid
)
956 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
959 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
960 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
962 if (key
.offset
<= start
&& extent_end
> end
) {
963 *length
= end
- start
+ 1;
965 } else if (key
.offset
<= start
&& extent_end
> start
)
966 *length
+= extent_end
- start
;
967 else if (key
.offset
> start
&& extent_end
<= end
)
968 *length
+= extent_end
- key
.offset
;
969 else if (key
.offset
> start
&& key
.offset
<= end
) {
970 *length
+= end
- key
.offset
+ 1;
972 } else if (key
.offset
> end
)
980 btrfs_free_path(path
);
985 * find_free_dev_extent - find free space in the specified device
986 * @device: the device which we search the free space in
987 * @num_bytes: the size of the free space that we need
988 * @start: store the start of the free space.
989 * @len: the size of the free space. that we find, or the size of the max
990 * free space if we don't find suitable free space
992 * this uses a pretty simple search, the expectation is that it is
993 * called very infrequently and that a given device has a small number
996 * @start is used to store the start of the free space if we find. But if we
997 * don't find suitable free space, it will be used to store the start position
998 * of the max free space.
1000 * @len is used to store the size of the free space that we find.
1001 * But if we don't find suitable free space, it is used to store the size of
1002 * the max free space.
1004 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
1005 u64
*start
, u64
*len
)
1007 struct btrfs_key key
;
1008 struct btrfs_root
*root
= device
->dev_root
;
1009 struct btrfs_dev_extent
*dev_extent
;
1010 struct btrfs_path
*path
;
1016 u64 search_end
= device
->total_bytes
;
1019 struct extent_buffer
*l
;
1021 /* FIXME use last free of some kind */
1023 /* we don't want to overwrite the superblock on the drive,
1024 * so we make sure to start at an offset of at least 1MB
1026 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1028 max_hole_start
= search_start
;
1032 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1037 path
= btrfs_alloc_path();
1044 key
.objectid
= device
->devid
;
1045 key
.offset
= search_start
;
1046 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1048 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1052 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1059 slot
= path
->slots
[0];
1060 if (slot
>= btrfs_header_nritems(l
)) {
1061 ret
= btrfs_next_leaf(root
, path
);
1069 btrfs_item_key_to_cpu(l
, &key
, slot
);
1071 if (key
.objectid
< device
->devid
)
1074 if (key
.objectid
> device
->devid
)
1077 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
1080 if (key
.offset
> search_start
) {
1081 hole_size
= key
.offset
- search_start
;
1083 if (hole_size
> max_hole_size
) {
1084 max_hole_start
= search_start
;
1085 max_hole_size
= hole_size
;
1089 * If this free space is greater than which we need,
1090 * it must be the max free space that we have found
1091 * until now, so max_hole_start must point to the start
1092 * of this free space and the length of this free space
1093 * is stored in max_hole_size. Thus, we return
1094 * max_hole_start and max_hole_size and go back to the
1097 if (hole_size
>= num_bytes
) {
1103 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1104 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1106 if (extent_end
> search_start
)
1107 search_start
= extent_end
;
1114 * At this point, search_start should be the end of
1115 * allocated dev extents, and when shrinking the device,
1116 * search_end may be smaller than search_start.
1118 if (search_end
> search_start
)
1119 hole_size
= search_end
- search_start
;
1121 if (hole_size
> max_hole_size
) {
1122 max_hole_start
= search_start
;
1123 max_hole_size
= hole_size
;
1127 if (hole_size
< num_bytes
)
1133 btrfs_free_path(path
);
1135 *start
= max_hole_start
;
1137 *len
= max_hole_size
;
1141 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1142 struct btrfs_device
*device
,
1146 struct btrfs_path
*path
;
1147 struct btrfs_root
*root
= device
->dev_root
;
1148 struct btrfs_key key
;
1149 struct btrfs_key found_key
;
1150 struct extent_buffer
*leaf
= NULL
;
1151 struct btrfs_dev_extent
*extent
= NULL
;
1153 path
= btrfs_alloc_path();
1157 key
.objectid
= device
->devid
;
1159 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1161 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1163 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1164 BTRFS_DEV_EXTENT_KEY
);
1167 leaf
= path
->nodes
[0];
1168 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1169 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1170 struct btrfs_dev_extent
);
1171 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1172 btrfs_dev_extent_length(leaf
, extent
) < start
);
1174 btrfs_release_path(path
);
1176 } else if (ret
== 0) {
1177 leaf
= path
->nodes
[0];
1178 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1179 struct btrfs_dev_extent
);
1181 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1185 if (device
->bytes_used
> 0) {
1186 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1187 device
->bytes_used
-= len
;
1188 spin_lock(&root
->fs_info
->free_chunk_lock
);
1189 root
->fs_info
->free_chunk_space
+= len
;
1190 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1192 ret
= btrfs_del_item(trans
, root
, path
);
1194 btrfs_error(root
->fs_info
, ret
,
1195 "Failed to remove dev extent item");
1198 btrfs_free_path(path
);
1202 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1203 struct btrfs_device
*device
,
1204 u64 chunk_tree
, u64 chunk_objectid
,
1205 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1208 struct btrfs_path
*path
;
1209 struct btrfs_root
*root
= device
->dev_root
;
1210 struct btrfs_dev_extent
*extent
;
1211 struct extent_buffer
*leaf
;
1212 struct btrfs_key key
;
1214 WARN_ON(!device
->in_fs_metadata
);
1215 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1216 path
= btrfs_alloc_path();
1220 key
.objectid
= device
->devid
;
1222 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1223 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1228 leaf
= path
->nodes
[0];
1229 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1230 struct btrfs_dev_extent
);
1231 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1232 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1233 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1235 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1236 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1239 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1240 btrfs_mark_buffer_dirty(leaf
);
1242 btrfs_free_path(path
);
1246 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1247 u64 objectid
, u64
*offset
)
1249 struct btrfs_path
*path
;
1251 struct btrfs_key key
;
1252 struct btrfs_chunk
*chunk
;
1253 struct btrfs_key found_key
;
1255 path
= btrfs_alloc_path();
1259 key
.objectid
= objectid
;
1260 key
.offset
= (u64
)-1;
1261 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1263 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1267 BUG_ON(ret
== 0); /* Corruption */
1269 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1273 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1275 if (found_key
.objectid
!= objectid
)
1278 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1279 struct btrfs_chunk
);
1280 *offset
= found_key
.offset
+
1281 btrfs_chunk_length(path
->nodes
[0], chunk
);
1286 btrfs_free_path(path
);
1290 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1293 struct btrfs_key key
;
1294 struct btrfs_key found_key
;
1295 struct btrfs_path
*path
;
1297 root
= root
->fs_info
->chunk_root
;
1299 path
= btrfs_alloc_path();
1303 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1304 key
.type
= BTRFS_DEV_ITEM_KEY
;
1305 key
.offset
= (u64
)-1;
1307 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1311 BUG_ON(ret
== 0); /* Corruption */
1313 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1314 BTRFS_DEV_ITEM_KEY
);
1318 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1320 *objectid
= found_key
.offset
+ 1;
1324 btrfs_free_path(path
);
1329 * the device information is stored in the chunk root
1330 * the btrfs_device struct should be fully filled in
1332 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1333 struct btrfs_root
*root
,
1334 struct btrfs_device
*device
)
1337 struct btrfs_path
*path
;
1338 struct btrfs_dev_item
*dev_item
;
1339 struct extent_buffer
*leaf
;
1340 struct btrfs_key key
;
1343 root
= root
->fs_info
->chunk_root
;
1345 path
= btrfs_alloc_path();
1349 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1350 key
.type
= BTRFS_DEV_ITEM_KEY
;
1351 key
.offset
= device
->devid
;
1353 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1358 leaf
= path
->nodes
[0];
1359 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1361 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1362 btrfs_set_device_generation(leaf
, dev_item
, 0);
1363 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1364 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1365 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1366 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1367 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1368 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1369 btrfs_set_device_group(leaf
, dev_item
, 0);
1370 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1371 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1372 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1374 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1375 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1376 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1377 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1378 btrfs_mark_buffer_dirty(leaf
);
1382 btrfs_free_path(path
);
1386 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1387 struct btrfs_device
*device
)
1390 struct btrfs_path
*path
;
1391 struct btrfs_key key
;
1392 struct btrfs_trans_handle
*trans
;
1394 root
= root
->fs_info
->chunk_root
;
1396 path
= btrfs_alloc_path();
1400 trans
= btrfs_start_transaction(root
, 0);
1401 if (IS_ERR(trans
)) {
1402 btrfs_free_path(path
);
1403 return PTR_ERR(trans
);
1405 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1406 key
.type
= BTRFS_DEV_ITEM_KEY
;
1407 key
.offset
= device
->devid
;
1410 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1419 ret
= btrfs_del_item(trans
, root
, path
);
1423 btrfs_free_path(path
);
1424 unlock_chunks(root
);
1425 btrfs_commit_transaction(trans
, root
);
1429 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1431 struct btrfs_device
*device
;
1432 struct btrfs_device
*next_device
;
1433 struct block_device
*bdev
;
1434 struct buffer_head
*bh
= NULL
;
1435 struct btrfs_super_block
*disk_super
;
1436 struct btrfs_fs_devices
*cur_devices
;
1443 bool clear_super
= false;
1445 mutex_lock(&uuid_mutex
);
1448 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1450 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1451 root
->fs_info
->avail_system_alloc_bits
|
1452 root
->fs_info
->avail_metadata_alloc_bits
;
1453 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1455 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1456 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1457 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1458 WARN_ON(num_devices
< 1);
1461 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1463 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1464 printk(KERN_ERR
"btrfs: unable to go below four devices "
1470 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1471 printk(KERN_ERR
"btrfs: unable to go below two "
1472 "devices on raid1\n");
1477 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1478 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1479 printk(KERN_ERR
"btrfs: unable to go below two "
1480 "devices on raid5\n");
1484 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1485 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1486 printk(KERN_ERR
"btrfs: unable to go below three "
1487 "devices on raid6\n");
1492 if (strcmp(device_path
, "missing") == 0) {
1493 struct list_head
*devices
;
1494 struct btrfs_device
*tmp
;
1497 devices
= &root
->fs_info
->fs_devices
->devices
;
1499 * It is safe to read the devices since the volume_mutex
1502 list_for_each_entry(tmp
, devices
, dev_list
) {
1503 if (tmp
->in_fs_metadata
&&
1504 !tmp
->is_tgtdev_for_dev_replace
&&
1514 printk(KERN_ERR
"btrfs: no missing devices found to "
1519 ret
= btrfs_get_bdev_and_sb(device_path
,
1520 FMODE_WRITE
| FMODE_EXCL
,
1521 root
->fs_info
->bdev_holder
, 0,
1525 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1526 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1527 dev_uuid
= disk_super
->dev_item
.uuid
;
1528 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1536 if (device
->is_tgtdev_for_dev_replace
) {
1537 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1542 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1543 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1549 if (device
->writeable
) {
1551 list_del_init(&device
->dev_alloc_list
);
1552 unlock_chunks(root
);
1553 root
->fs_info
->fs_devices
->rw_devices
--;
1557 ret
= btrfs_shrink_device(device
, 0);
1562 * TODO: the superblock still includes this device in its num_devices
1563 * counter although write_all_supers() is not locked out. This
1564 * could give a filesystem state which requires a degraded mount.
1566 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1570 spin_lock(&root
->fs_info
->free_chunk_lock
);
1571 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1573 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1575 device
->in_fs_metadata
= 0;
1576 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1579 * the device list mutex makes sure that we don't change
1580 * the device list while someone else is writing out all
1581 * the device supers.
1584 cur_devices
= device
->fs_devices
;
1585 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1586 list_del_rcu(&device
->dev_list
);
1588 device
->fs_devices
->num_devices
--;
1589 device
->fs_devices
->total_devices
--;
1591 if (device
->missing
)
1592 root
->fs_info
->fs_devices
->missing_devices
--;
1594 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1595 struct btrfs_device
, dev_list
);
1596 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1597 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1598 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1599 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1602 device
->fs_devices
->open_devices
--;
1604 call_rcu(&device
->rcu
, free_device
);
1605 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1607 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1608 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1610 if (cur_devices
->open_devices
== 0) {
1611 struct btrfs_fs_devices
*fs_devices
;
1612 fs_devices
= root
->fs_info
->fs_devices
;
1613 while (fs_devices
) {
1614 if (fs_devices
->seed
== cur_devices
)
1616 fs_devices
= fs_devices
->seed
;
1618 fs_devices
->seed
= cur_devices
->seed
;
1619 cur_devices
->seed
= NULL
;
1621 __btrfs_close_devices(cur_devices
);
1622 unlock_chunks(root
);
1623 free_fs_devices(cur_devices
);
1626 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1627 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1630 * at this point, the device is zero sized. We want to
1631 * remove it from the devices list and zero out the old super
1633 if (clear_super
&& disk_super
) {
1634 /* make sure this device isn't detected as part of
1637 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1638 set_buffer_dirty(bh
);
1639 sync_dirty_buffer(bh
);
1644 /* Notify udev that device has changed */
1646 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1651 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1653 mutex_unlock(&uuid_mutex
);
1656 if (device
->writeable
) {
1658 list_add(&device
->dev_alloc_list
,
1659 &root
->fs_info
->fs_devices
->alloc_list
);
1660 unlock_chunks(root
);
1661 root
->fs_info
->fs_devices
->rw_devices
++;
1666 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1667 struct btrfs_device
*srcdev
)
1669 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1670 list_del_rcu(&srcdev
->dev_list
);
1671 list_del_rcu(&srcdev
->dev_alloc_list
);
1672 fs_info
->fs_devices
->num_devices
--;
1673 if (srcdev
->missing
) {
1674 fs_info
->fs_devices
->missing_devices
--;
1675 fs_info
->fs_devices
->rw_devices
++;
1677 if (srcdev
->can_discard
)
1678 fs_info
->fs_devices
->num_can_discard
--;
1680 fs_info
->fs_devices
->open_devices
--;
1682 call_rcu(&srcdev
->rcu
, free_device
);
1685 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1686 struct btrfs_device
*tgtdev
)
1688 struct btrfs_device
*next_device
;
1691 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1693 btrfs_scratch_superblock(tgtdev
);
1694 fs_info
->fs_devices
->open_devices
--;
1696 fs_info
->fs_devices
->num_devices
--;
1697 if (tgtdev
->can_discard
)
1698 fs_info
->fs_devices
->num_can_discard
++;
1700 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1701 struct btrfs_device
, dev_list
);
1702 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1703 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1704 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1705 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1706 list_del_rcu(&tgtdev
->dev_list
);
1708 call_rcu(&tgtdev
->rcu
, free_device
);
1710 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1713 int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1714 struct btrfs_device
**device
)
1717 struct btrfs_super_block
*disk_super
;
1720 struct block_device
*bdev
;
1721 struct buffer_head
*bh
;
1724 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1725 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1728 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1729 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1730 dev_uuid
= disk_super
->dev_item
.uuid
;
1731 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1736 blkdev_put(bdev
, FMODE_READ
);
1740 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1742 struct btrfs_device
**device
)
1745 if (strcmp(device_path
, "missing") == 0) {
1746 struct list_head
*devices
;
1747 struct btrfs_device
*tmp
;
1749 devices
= &root
->fs_info
->fs_devices
->devices
;
1751 * It is safe to read the devices since the volume_mutex
1752 * is held by the caller.
1754 list_for_each_entry(tmp
, devices
, dev_list
) {
1755 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1762 pr_err("btrfs: no missing device found\n");
1768 return btrfs_find_device_by_path(root
, device_path
, device
);
1773 * does all the dirty work required for changing file system's UUID.
1775 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1777 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1778 struct btrfs_fs_devices
*old_devices
;
1779 struct btrfs_fs_devices
*seed_devices
;
1780 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1781 struct btrfs_device
*device
;
1784 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1785 if (!fs_devices
->seeding
)
1788 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1792 old_devices
= clone_fs_devices(fs_devices
);
1793 if (IS_ERR(old_devices
)) {
1794 kfree(seed_devices
);
1795 return PTR_ERR(old_devices
);
1798 list_add(&old_devices
->list
, &fs_uuids
);
1800 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1801 seed_devices
->opened
= 1;
1802 INIT_LIST_HEAD(&seed_devices
->devices
);
1803 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1804 mutex_init(&seed_devices
->device_list_mutex
);
1806 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1807 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1809 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1811 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1812 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1813 device
->fs_devices
= seed_devices
;
1816 fs_devices
->seeding
= 0;
1817 fs_devices
->num_devices
= 0;
1818 fs_devices
->open_devices
= 0;
1819 fs_devices
->total_devices
= 0;
1820 fs_devices
->seed
= seed_devices
;
1822 generate_random_uuid(fs_devices
->fsid
);
1823 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1824 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1825 super_flags
= btrfs_super_flags(disk_super
) &
1826 ~BTRFS_SUPER_FLAG_SEEDING
;
1827 btrfs_set_super_flags(disk_super
, super_flags
);
1833 * strore the expected generation for seed devices in device items.
1835 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1836 struct btrfs_root
*root
)
1838 struct btrfs_path
*path
;
1839 struct extent_buffer
*leaf
;
1840 struct btrfs_dev_item
*dev_item
;
1841 struct btrfs_device
*device
;
1842 struct btrfs_key key
;
1843 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1844 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1848 path
= btrfs_alloc_path();
1852 root
= root
->fs_info
->chunk_root
;
1853 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1855 key
.type
= BTRFS_DEV_ITEM_KEY
;
1858 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1862 leaf
= path
->nodes
[0];
1864 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1865 ret
= btrfs_next_leaf(root
, path
);
1870 leaf
= path
->nodes
[0];
1871 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1872 btrfs_release_path(path
);
1876 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1877 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1878 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1881 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1882 struct btrfs_dev_item
);
1883 devid
= btrfs_device_id(leaf
, dev_item
);
1884 read_extent_buffer(leaf
, dev_uuid
,
1885 (unsigned long)btrfs_device_uuid(dev_item
),
1887 read_extent_buffer(leaf
, fs_uuid
,
1888 (unsigned long)btrfs_device_fsid(dev_item
),
1890 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1892 BUG_ON(!device
); /* Logic error */
1894 if (device
->fs_devices
->seeding
) {
1895 btrfs_set_device_generation(leaf
, dev_item
,
1896 device
->generation
);
1897 btrfs_mark_buffer_dirty(leaf
);
1905 btrfs_free_path(path
);
1909 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1911 struct request_queue
*q
;
1912 struct btrfs_trans_handle
*trans
;
1913 struct btrfs_device
*device
;
1914 struct block_device
*bdev
;
1915 struct list_head
*devices
;
1916 struct super_block
*sb
= root
->fs_info
->sb
;
1917 struct rcu_string
*name
;
1919 int seeding_dev
= 0;
1922 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1925 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1926 root
->fs_info
->bdev_holder
);
1928 return PTR_ERR(bdev
);
1930 if (root
->fs_info
->fs_devices
->seeding
) {
1932 down_write(&sb
->s_umount
);
1933 mutex_lock(&uuid_mutex
);
1936 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1938 devices
= &root
->fs_info
->fs_devices
->devices
;
1940 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1941 list_for_each_entry(device
, devices
, dev_list
) {
1942 if (device
->bdev
== bdev
) {
1945 &root
->fs_info
->fs_devices
->device_list_mutex
);
1949 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1951 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1953 /* we can safely leave the fs_devices entry around */
1958 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
1964 rcu_assign_pointer(device
->name
, name
);
1966 ret
= find_next_devid(root
, &device
->devid
);
1968 rcu_string_free(device
->name
);
1973 trans
= btrfs_start_transaction(root
, 0);
1974 if (IS_ERR(trans
)) {
1975 rcu_string_free(device
->name
);
1977 ret
= PTR_ERR(trans
);
1983 q
= bdev_get_queue(bdev
);
1984 if (blk_queue_discard(q
))
1985 device
->can_discard
= 1;
1986 device
->writeable
= 1;
1987 device
->work
.func
= pending_bios_fn
;
1988 generate_random_uuid(device
->uuid
);
1989 spin_lock_init(&device
->io_lock
);
1990 device
->generation
= trans
->transid
;
1991 device
->io_width
= root
->sectorsize
;
1992 device
->io_align
= root
->sectorsize
;
1993 device
->sector_size
= root
->sectorsize
;
1994 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1995 device
->disk_total_bytes
= device
->total_bytes
;
1996 device
->dev_root
= root
->fs_info
->dev_root
;
1997 device
->bdev
= bdev
;
1998 device
->in_fs_metadata
= 1;
1999 device
->is_tgtdev_for_dev_replace
= 0;
2000 device
->mode
= FMODE_EXCL
;
2001 set_blocksize(device
->bdev
, 4096);
2004 sb
->s_flags
&= ~MS_RDONLY
;
2005 ret
= btrfs_prepare_sprout(root
);
2006 BUG_ON(ret
); /* -ENOMEM */
2009 device
->fs_devices
= root
->fs_info
->fs_devices
;
2011 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2012 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2013 list_add(&device
->dev_alloc_list
,
2014 &root
->fs_info
->fs_devices
->alloc_list
);
2015 root
->fs_info
->fs_devices
->num_devices
++;
2016 root
->fs_info
->fs_devices
->open_devices
++;
2017 root
->fs_info
->fs_devices
->rw_devices
++;
2018 root
->fs_info
->fs_devices
->total_devices
++;
2019 if (device
->can_discard
)
2020 root
->fs_info
->fs_devices
->num_can_discard
++;
2021 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2023 spin_lock(&root
->fs_info
->free_chunk_lock
);
2024 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2025 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2027 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2028 root
->fs_info
->fs_devices
->rotating
= 1;
2030 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2031 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2032 total_bytes
+ device
->total_bytes
);
2034 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2035 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2037 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2040 ret
= init_first_rw_device(trans
, root
, device
);
2042 btrfs_abort_transaction(trans
, root
, ret
);
2045 ret
= btrfs_finish_sprout(trans
, root
);
2047 btrfs_abort_transaction(trans
, root
, ret
);
2051 ret
= btrfs_add_device(trans
, root
, device
);
2053 btrfs_abort_transaction(trans
, root
, ret
);
2059 * we've got more storage, clear any full flags on the space
2062 btrfs_clear_space_info_full(root
->fs_info
);
2064 unlock_chunks(root
);
2065 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2066 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2067 ret
= btrfs_commit_transaction(trans
, root
);
2070 mutex_unlock(&uuid_mutex
);
2071 up_write(&sb
->s_umount
);
2073 if (ret
) /* transaction commit */
2076 ret
= btrfs_relocate_sys_chunks(root
);
2078 btrfs_error(root
->fs_info
, ret
,
2079 "Failed to relocate sys chunks after "
2080 "device initialization. This can be fixed "
2081 "using the \"btrfs balance\" command.");
2082 trans
= btrfs_attach_transaction(root
);
2083 if (IS_ERR(trans
)) {
2084 if (PTR_ERR(trans
) == -ENOENT
)
2086 return PTR_ERR(trans
);
2088 ret
= btrfs_commit_transaction(trans
, root
);
2094 unlock_chunks(root
);
2095 btrfs_end_transaction(trans
, root
);
2096 rcu_string_free(device
->name
);
2099 blkdev_put(bdev
, FMODE_EXCL
);
2101 mutex_unlock(&uuid_mutex
);
2102 up_write(&sb
->s_umount
);
2107 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2108 struct btrfs_device
**device_out
)
2110 struct request_queue
*q
;
2111 struct btrfs_device
*device
;
2112 struct block_device
*bdev
;
2113 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2114 struct list_head
*devices
;
2115 struct rcu_string
*name
;
2119 if (fs_info
->fs_devices
->seeding
)
2122 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2123 fs_info
->bdev_holder
);
2125 return PTR_ERR(bdev
);
2127 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2129 devices
= &fs_info
->fs_devices
->devices
;
2130 list_for_each_entry(device
, devices
, dev_list
) {
2131 if (device
->bdev
== bdev
) {
2137 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
2143 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2149 rcu_assign_pointer(device
->name
, name
);
2151 q
= bdev_get_queue(bdev
);
2152 if (blk_queue_discard(q
))
2153 device
->can_discard
= 1;
2154 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2155 device
->writeable
= 1;
2156 device
->work
.func
= pending_bios_fn
;
2157 generate_random_uuid(device
->uuid
);
2158 device
->devid
= BTRFS_DEV_REPLACE_DEVID
;
2159 spin_lock_init(&device
->io_lock
);
2160 device
->generation
= 0;
2161 device
->io_width
= root
->sectorsize
;
2162 device
->io_align
= root
->sectorsize
;
2163 device
->sector_size
= root
->sectorsize
;
2164 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2165 device
->disk_total_bytes
= device
->total_bytes
;
2166 device
->dev_root
= fs_info
->dev_root
;
2167 device
->bdev
= bdev
;
2168 device
->in_fs_metadata
= 1;
2169 device
->is_tgtdev_for_dev_replace
= 1;
2170 device
->mode
= FMODE_EXCL
;
2171 set_blocksize(device
->bdev
, 4096);
2172 device
->fs_devices
= fs_info
->fs_devices
;
2173 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2174 fs_info
->fs_devices
->num_devices
++;
2175 fs_info
->fs_devices
->open_devices
++;
2176 if (device
->can_discard
)
2177 fs_info
->fs_devices
->num_can_discard
++;
2178 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2180 *device_out
= device
;
2184 blkdev_put(bdev
, FMODE_EXCL
);
2188 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2189 struct btrfs_device
*tgtdev
)
2191 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2192 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2193 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2194 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2195 tgtdev
->dev_root
= fs_info
->dev_root
;
2196 tgtdev
->in_fs_metadata
= 1;
2199 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2200 struct btrfs_device
*device
)
2203 struct btrfs_path
*path
;
2204 struct btrfs_root
*root
;
2205 struct btrfs_dev_item
*dev_item
;
2206 struct extent_buffer
*leaf
;
2207 struct btrfs_key key
;
2209 root
= device
->dev_root
->fs_info
->chunk_root
;
2211 path
= btrfs_alloc_path();
2215 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2216 key
.type
= BTRFS_DEV_ITEM_KEY
;
2217 key
.offset
= device
->devid
;
2219 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2228 leaf
= path
->nodes
[0];
2229 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2231 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2232 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2233 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2234 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2235 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2236 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
2237 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
2238 btrfs_mark_buffer_dirty(leaf
);
2241 btrfs_free_path(path
);
2245 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2246 struct btrfs_device
*device
, u64 new_size
)
2248 struct btrfs_super_block
*super_copy
=
2249 device
->dev_root
->fs_info
->super_copy
;
2250 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2251 u64 diff
= new_size
- device
->total_bytes
;
2253 if (!device
->writeable
)
2255 if (new_size
<= device
->total_bytes
||
2256 device
->is_tgtdev_for_dev_replace
)
2259 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2260 device
->fs_devices
->total_rw_bytes
+= diff
;
2262 device
->total_bytes
= new_size
;
2263 device
->disk_total_bytes
= new_size
;
2264 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2266 return btrfs_update_device(trans
, device
);
2269 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2270 struct btrfs_device
*device
, u64 new_size
)
2273 lock_chunks(device
->dev_root
);
2274 ret
= __btrfs_grow_device(trans
, device
, new_size
);
2275 unlock_chunks(device
->dev_root
);
2279 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2280 struct btrfs_root
*root
,
2281 u64 chunk_tree
, u64 chunk_objectid
,
2285 struct btrfs_path
*path
;
2286 struct btrfs_key key
;
2288 root
= root
->fs_info
->chunk_root
;
2289 path
= btrfs_alloc_path();
2293 key
.objectid
= chunk_objectid
;
2294 key
.offset
= chunk_offset
;
2295 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2297 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2300 else if (ret
> 0) { /* Logic error or corruption */
2301 btrfs_error(root
->fs_info
, -ENOENT
,
2302 "Failed lookup while freeing chunk.");
2307 ret
= btrfs_del_item(trans
, root
, path
);
2309 btrfs_error(root
->fs_info
, ret
,
2310 "Failed to delete chunk item.");
2312 btrfs_free_path(path
);
2316 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2319 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2320 struct btrfs_disk_key
*disk_key
;
2321 struct btrfs_chunk
*chunk
;
2328 struct btrfs_key key
;
2330 array_size
= btrfs_super_sys_array_size(super_copy
);
2332 ptr
= super_copy
->sys_chunk_array
;
2335 while (cur
< array_size
) {
2336 disk_key
= (struct btrfs_disk_key
*)ptr
;
2337 btrfs_disk_key_to_cpu(&key
, disk_key
);
2339 len
= sizeof(*disk_key
);
2341 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2342 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2343 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2344 len
+= btrfs_chunk_item_size(num_stripes
);
2349 if (key
.objectid
== chunk_objectid
&&
2350 key
.offset
== chunk_offset
) {
2351 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2353 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2362 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2363 u64 chunk_tree
, u64 chunk_objectid
,
2366 struct extent_map_tree
*em_tree
;
2367 struct btrfs_root
*extent_root
;
2368 struct btrfs_trans_handle
*trans
;
2369 struct extent_map
*em
;
2370 struct map_lookup
*map
;
2374 root
= root
->fs_info
->chunk_root
;
2375 extent_root
= root
->fs_info
->extent_root
;
2376 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2378 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2382 /* step one, relocate all the extents inside this chunk */
2383 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2387 trans
= btrfs_start_transaction(root
, 0);
2388 if (IS_ERR(trans
)) {
2389 ret
= PTR_ERR(trans
);
2390 btrfs_std_error(root
->fs_info
, ret
);
2397 * step two, delete the device extents and the
2398 * chunk tree entries
2400 read_lock(&em_tree
->lock
);
2401 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2402 read_unlock(&em_tree
->lock
);
2404 BUG_ON(!em
|| em
->start
> chunk_offset
||
2405 em
->start
+ em
->len
< chunk_offset
);
2406 map
= (struct map_lookup
*)em
->bdev
;
2408 for (i
= 0; i
< map
->num_stripes
; i
++) {
2409 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2410 map
->stripes
[i
].physical
);
2413 if (map
->stripes
[i
].dev
) {
2414 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2418 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2423 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2425 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2426 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2430 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2433 write_lock(&em_tree
->lock
);
2434 remove_extent_mapping(em_tree
, em
);
2435 write_unlock(&em_tree
->lock
);
2440 /* once for the tree */
2441 free_extent_map(em
);
2443 free_extent_map(em
);
2445 unlock_chunks(root
);
2446 btrfs_end_transaction(trans
, root
);
2450 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2452 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2453 struct btrfs_path
*path
;
2454 struct extent_buffer
*leaf
;
2455 struct btrfs_chunk
*chunk
;
2456 struct btrfs_key key
;
2457 struct btrfs_key found_key
;
2458 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2460 bool retried
= false;
2464 path
= btrfs_alloc_path();
2469 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2470 key
.offset
= (u64
)-1;
2471 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2474 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2477 BUG_ON(ret
== 0); /* Corruption */
2479 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2486 leaf
= path
->nodes
[0];
2487 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2489 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2490 struct btrfs_chunk
);
2491 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2492 btrfs_release_path(path
);
2494 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2495 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2504 if (found_key
.offset
== 0)
2506 key
.offset
= found_key
.offset
- 1;
2509 if (failed
&& !retried
) {
2513 } else if (failed
&& retried
) {
2518 btrfs_free_path(path
);
2522 static int insert_balance_item(struct btrfs_root
*root
,
2523 struct btrfs_balance_control
*bctl
)
2525 struct btrfs_trans_handle
*trans
;
2526 struct btrfs_balance_item
*item
;
2527 struct btrfs_disk_balance_args disk_bargs
;
2528 struct btrfs_path
*path
;
2529 struct extent_buffer
*leaf
;
2530 struct btrfs_key key
;
2533 path
= btrfs_alloc_path();
2537 trans
= btrfs_start_transaction(root
, 0);
2538 if (IS_ERR(trans
)) {
2539 btrfs_free_path(path
);
2540 return PTR_ERR(trans
);
2543 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2544 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2547 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2552 leaf
= path
->nodes
[0];
2553 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2555 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2557 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2558 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2559 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2560 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2561 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2562 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2564 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2566 btrfs_mark_buffer_dirty(leaf
);
2568 btrfs_free_path(path
);
2569 err
= btrfs_commit_transaction(trans
, root
);
2575 static int del_balance_item(struct btrfs_root
*root
)
2577 struct btrfs_trans_handle
*trans
;
2578 struct btrfs_path
*path
;
2579 struct btrfs_key key
;
2582 path
= btrfs_alloc_path();
2586 trans
= btrfs_start_transaction(root
, 0);
2587 if (IS_ERR(trans
)) {
2588 btrfs_free_path(path
);
2589 return PTR_ERR(trans
);
2592 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2593 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2596 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2604 ret
= btrfs_del_item(trans
, root
, path
);
2606 btrfs_free_path(path
);
2607 err
= btrfs_commit_transaction(trans
, root
);
2614 * This is a heuristic used to reduce the number of chunks balanced on
2615 * resume after balance was interrupted.
2617 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2620 * Turn on soft mode for chunk types that were being converted.
2622 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2623 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2624 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2625 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2626 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2627 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2630 * Turn on usage filter if is not already used. The idea is
2631 * that chunks that we have already balanced should be
2632 * reasonably full. Don't do it for chunks that are being
2633 * converted - that will keep us from relocating unconverted
2634 * (albeit full) chunks.
2636 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2637 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2638 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2639 bctl
->data
.usage
= 90;
2641 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2642 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2643 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2644 bctl
->sys
.usage
= 90;
2646 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2647 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2648 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2649 bctl
->meta
.usage
= 90;
2654 * Should be called with both balance and volume mutexes held to
2655 * serialize other volume operations (add_dev/rm_dev/resize) with
2656 * restriper. Same goes for unset_balance_control.
2658 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2660 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2662 BUG_ON(fs_info
->balance_ctl
);
2664 spin_lock(&fs_info
->balance_lock
);
2665 fs_info
->balance_ctl
= bctl
;
2666 spin_unlock(&fs_info
->balance_lock
);
2669 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2671 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2673 BUG_ON(!fs_info
->balance_ctl
);
2675 spin_lock(&fs_info
->balance_lock
);
2676 fs_info
->balance_ctl
= NULL
;
2677 spin_unlock(&fs_info
->balance_lock
);
2683 * Balance filters. Return 1 if chunk should be filtered out
2684 * (should not be balanced).
2686 static int chunk_profiles_filter(u64 chunk_type
,
2687 struct btrfs_balance_args
*bargs
)
2689 chunk_type
= chunk_to_extended(chunk_type
) &
2690 BTRFS_EXTENDED_PROFILE_MASK
;
2692 if (bargs
->profiles
& chunk_type
)
2698 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2699 struct btrfs_balance_args
*bargs
)
2701 struct btrfs_block_group_cache
*cache
;
2702 u64 chunk_used
, user_thresh
;
2705 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2706 chunk_used
= btrfs_block_group_used(&cache
->item
);
2708 if (bargs
->usage
== 0)
2710 else if (bargs
->usage
> 100)
2711 user_thresh
= cache
->key
.offset
;
2713 user_thresh
= div_factor_fine(cache
->key
.offset
,
2716 if (chunk_used
< user_thresh
)
2719 btrfs_put_block_group(cache
);
2723 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2724 struct btrfs_chunk
*chunk
,
2725 struct btrfs_balance_args
*bargs
)
2727 struct btrfs_stripe
*stripe
;
2728 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2731 for (i
= 0; i
< num_stripes
; i
++) {
2732 stripe
= btrfs_stripe_nr(chunk
, i
);
2733 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2740 /* [pstart, pend) */
2741 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2742 struct btrfs_chunk
*chunk
,
2744 struct btrfs_balance_args
*bargs
)
2746 struct btrfs_stripe
*stripe
;
2747 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2753 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2756 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2757 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
2758 factor
= num_stripes
/ 2;
2759 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
2760 factor
= num_stripes
- 1;
2761 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
2762 factor
= num_stripes
- 2;
2764 factor
= num_stripes
;
2767 for (i
= 0; i
< num_stripes
; i
++) {
2768 stripe
= btrfs_stripe_nr(chunk
, i
);
2769 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2772 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2773 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2774 do_div(stripe_length
, factor
);
2776 if (stripe_offset
< bargs
->pend
&&
2777 stripe_offset
+ stripe_length
> bargs
->pstart
)
2784 /* [vstart, vend) */
2785 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2786 struct btrfs_chunk
*chunk
,
2788 struct btrfs_balance_args
*bargs
)
2790 if (chunk_offset
< bargs
->vend
&&
2791 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2792 /* at least part of the chunk is inside this vrange */
2798 static int chunk_soft_convert_filter(u64 chunk_type
,
2799 struct btrfs_balance_args
*bargs
)
2801 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2804 chunk_type
= chunk_to_extended(chunk_type
) &
2805 BTRFS_EXTENDED_PROFILE_MASK
;
2807 if (bargs
->target
== chunk_type
)
2813 static int should_balance_chunk(struct btrfs_root
*root
,
2814 struct extent_buffer
*leaf
,
2815 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2817 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2818 struct btrfs_balance_args
*bargs
= NULL
;
2819 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2822 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2823 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2827 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2828 bargs
= &bctl
->data
;
2829 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2831 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2832 bargs
= &bctl
->meta
;
2834 /* profiles filter */
2835 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2836 chunk_profiles_filter(chunk_type
, bargs
)) {
2841 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2842 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2847 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2848 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2852 /* drange filter, makes sense only with devid filter */
2853 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2854 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2859 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2860 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2864 /* soft profile changing mode */
2865 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2866 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2873 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2875 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2876 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2877 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2878 struct list_head
*devices
;
2879 struct btrfs_device
*device
;
2882 struct btrfs_chunk
*chunk
;
2883 struct btrfs_path
*path
;
2884 struct btrfs_key key
;
2885 struct btrfs_key found_key
;
2886 struct btrfs_trans_handle
*trans
;
2887 struct extent_buffer
*leaf
;
2890 int enospc_errors
= 0;
2891 bool counting
= true;
2893 /* step one make some room on all the devices */
2894 devices
= &fs_info
->fs_devices
->devices
;
2895 list_for_each_entry(device
, devices
, dev_list
) {
2896 old_size
= device
->total_bytes
;
2897 size_to_free
= div_factor(old_size
, 1);
2898 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2899 if (!device
->writeable
||
2900 device
->total_bytes
- device
->bytes_used
> size_to_free
||
2901 device
->is_tgtdev_for_dev_replace
)
2904 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2909 trans
= btrfs_start_transaction(dev_root
, 0);
2910 BUG_ON(IS_ERR(trans
));
2912 ret
= btrfs_grow_device(trans
, device
, old_size
);
2915 btrfs_end_transaction(trans
, dev_root
);
2918 /* step two, relocate all the chunks */
2919 path
= btrfs_alloc_path();
2925 /* zero out stat counters */
2926 spin_lock(&fs_info
->balance_lock
);
2927 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2928 spin_unlock(&fs_info
->balance_lock
);
2930 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2931 key
.offset
= (u64
)-1;
2932 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2935 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2936 atomic_read(&fs_info
->balance_cancel_req
)) {
2941 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2946 * this shouldn't happen, it means the last relocate
2950 BUG(); /* FIXME break ? */
2952 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2953 BTRFS_CHUNK_ITEM_KEY
);
2959 leaf
= path
->nodes
[0];
2960 slot
= path
->slots
[0];
2961 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2963 if (found_key
.objectid
!= key
.objectid
)
2966 /* chunk zero is special */
2967 if (found_key
.offset
== 0)
2970 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2973 spin_lock(&fs_info
->balance_lock
);
2974 bctl
->stat
.considered
++;
2975 spin_unlock(&fs_info
->balance_lock
);
2978 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2980 btrfs_release_path(path
);
2985 spin_lock(&fs_info
->balance_lock
);
2986 bctl
->stat
.expected
++;
2987 spin_unlock(&fs_info
->balance_lock
);
2991 ret
= btrfs_relocate_chunk(chunk_root
,
2992 chunk_root
->root_key
.objectid
,
2995 if (ret
&& ret
!= -ENOSPC
)
2997 if (ret
== -ENOSPC
) {
3000 spin_lock(&fs_info
->balance_lock
);
3001 bctl
->stat
.completed
++;
3002 spin_unlock(&fs_info
->balance_lock
);
3005 key
.offset
= found_key
.offset
- 1;
3009 btrfs_release_path(path
);
3014 btrfs_free_path(path
);
3015 if (enospc_errors
) {
3016 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
3026 * alloc_profile_is_valid - see if a given profile is valid and reduced
3027 * @flags: profile to validate
3028 * @extended: if true @flags is treated as an extended profile
3030 static int alloc_profile_is_valid(u64 flags
, int extended
)
3032 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3033 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3035 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3037 /* 1) check that all other bits are zeroed */
3041 /* 2) see if profile is reduced */
3043 return !extended
; /* "0" is valid for usual profiles */
3045 /* true if exactly one bit set */
3046 return (flags
& (flags
- 1)) == 0;
3049 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3051 /* cancel requested || normal exit path */
3052 return atomic_read(&fs_info
->balance_cancel_req
) ||
3053 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3054 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3057 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3061 unset_balance_control(fs_info
);
3062 ret
= del_balance_item(fs_info
->tree_root
);
3064 btrfs_std_error(fs_info
, ret
);
3066 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3069 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
3070 struct btrfs_ioctl_balance_args
*bargs
);
3073 * Should be called with both balance and volume mutexes held
3075 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3076 struct btrfs_ioctl_balance_args
*bargs
)
3078 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3085 if (btrfs_fs_closing(fs_info
) ||
3086 atomic_read(&fs_info
->balance_pause_req
) ||
3087 atomic_read(&fs_info
->balance_cancel_req
)) {
3092 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3093 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3097 * In case of mixed groups both data and meta should be picked,
3098 * and identical options should be given for both of them.
3100 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3101 if (mixed
&& (bctl
->flags
& allowed
)) {
3102 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3103 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3104 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3105 printk(KERN_ERR
"btrfs: with mixed groups data and "
3106 "metadata balance options must be the same\n");
3112 num_devices
= fs_info
->fs_devices
->num_devices
;
3113 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3114 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3115 BUG_ON(num_devices
< 1);
3118 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3119 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3120 if (num_devices
== 1)
3121 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3122 else if (num_devices
< 4)
3123 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3125 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
3126 BTRFS_BLOCK_GROUP_RAID10
|
3127 BTRFS_BLOCK_GROUP_RAID5
|
3128 BTRFS_BLOCK_GROUP_RAID6
);
3130 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3131 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3132 (bctl
->data
.target
& ~allowed
))) {
3133 printk(KERN_ERR
"btrfs: unable to start balance with target "
3134 "data profile %llu\n",
3135 (unsigned long long)bctl
->data
.target
);
3139 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3140 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3141 (bctl
->meta
.target
& ~allowed
))) {
3142 printk(KERN_ERR
"btrfs: unable to start balance with target "
3143 "metadata profile %llu\n",
3144 (unsigned long long)bctl
->meta
.target
);
3148 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3149 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3150 (bctl
->sys
.target
& ~allowed
))) {
3151 printk(KERN_ERR
"btrfs: unable to start balance with target "
3152 "system profile %llu\n",
3153 (unsigned long long)bctl
->sys
.target
);
3158 /* allow dup'ed data chunks only in mixed mode */
3159 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3160 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3161 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
3166 /* allow to reduce meta or sys integrity only if force set */
3167 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3168 BTRFS_BLOCK_GROUP_RAID10
|
3169 BTRFS_BLOCK_GROUP_RAID5
|
3170 BTRFS_BLOCK_GROUP_RAID6
;
3172 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3174 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3175 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3176 !(bctl
->sys
.target
& allowed
)) ||
3177 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3178 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3179 !(bctl
->meta
.target
& allowed
))) {
3180 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3181 printk(KERN_INFO
"btrfs: force reducing metadata "
3184 printk(KERN_ERR
"btrfs: balance will reduce metadata "
3185 "integrity, use force if you want this\n");
3190 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3192 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3193 int num_tolerated_disk_barrier_failures
;
3194 u64 target
= bctl
->sys
.target
;
3196 num_tolerated_disk_barrier_failures
=
3197 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3198 if (num_tolerated_disk_barrier_failures
> 0 &&
3200 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3201 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3202 num_tolerated_disk_barrier_failures
= 0;
3203 else if (num_tolerated_disk_barrier_failures
> 1 &&
3205 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3206 num_tolerated_disk_barrier_failures
= 1;
3208 fs_info
->num_tolerated_disk_barrier_failures
=
3209 num_tolerated_disk_barrier_failures
;
3212 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3213 if (ret
&& ret
!= -EEXIST
)
3216 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3217 BUG_ON(ret
== -EEXIST
);
3218 set_balance_control(bctl
);
3220 BUG_ON(ret
!= -EEXIST
);
3221 spin_lock(&fs_info
->balance_lock
);
3222 update_balance_args(bctl
);
3223 spin_unlock(&fs_info
->balance_lock
);
3226 atomic_inc(&fs_info
->balance_running
);
3227 mutex_unlock(&fs_info
->balance_mutex
);
3229 ret
= __btrfs_balance(fs_info
);
3231 mutex_lock(&fs_info
->balance_mutex
);
3232 atomic_dec(&fs_info
->balance_running
);
3234 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3235 fs_info
->num_tolerated_disk_barrier_failures
=
3236 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3240 memset(bargs
, 0, sizeof(*bargs
));
3241 update_ioctl_balance_args(fs_info
, 0, bargs
);
3244 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3245 balance_need_close(fs_info
)) {
3246 __cancel_balance(fs_info
);
3249 wake_up(&fs_info
->balance_wait_q
);
3253 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3254 __cancel_balance(fs_info
);
3257 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3262 static int balance_kthread(void *data
)
3264 struct btrfs_fs_info
*fs_info
= data
;
3267 mutex_lock(&fs_info
->volume_mutex
);
3268 mutex_lock(&fs_info
->balance_mutex
);
3270 if (fs_info
->balance_ctl
) {
3271 printk(KERN_INFO
"btrfs: continuing balance\n");
3272 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3275 mutex_unlock(&fs_info
->balance_mutex
);
3276 mutex_unlock(&fs_info
->volume_mutex
);
3281 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3283 struct task_struct
*tsk
;
3285 spin_lock(&fs_info
->balance_lock
);
3286 if (!fs_info
->balance_ctl
) {
3287 spin_unlock(&fs_info
->balance_lock
);
3290 spin_unlock(&fs_info
->balance_lock
);
3292 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3293 printk(KERN_INFO
"btrfs: force skipping balance\n");
3297 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3299 return PTR_ERR(tsk
);
3304 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3306 struct btrfs_balance_control
*bctl
;
3307 struct btrfs_balance_item
*item
;
3308 struct btrfs_disk_balance_args disk_bargs
;
3309 struct btrfs_path
*path
;
3310 struct extent_buffer
*leaf
;
3311 struct btrfs_key key
;
3314 path
= btrfs_alloc_path();
3318 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3319 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3322 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3325 if (ret
> 0) { /* ret = -ENOENT; */
3330 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3336 leaf
= path
->nodes
[0];
3337 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3339 bctl
->fs_info
= fs_info
;
3340 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3341 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3343 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3344 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3345 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3346 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3347 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3348 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3350 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3352 mutex_lock(&fs_info
->volume_mutex
);
3353 mutex_lock(&fs_info
->balance_mutex
);
3355 set_balance_control(bctl
);
3357 mutex_unlock(&fs_info
->balance_mutex
);
3358 mutex_unlock(&fs_info
->volume_mutex
);
3360 btrfs_free_path(path
);
3364 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3368 mutex_lock(&fs_info
->balance_mutex
);
3369 if (!fs_info
->balance_ctl
) {
3370 mutex_unlock(&fs_info
->balance_mutex
);
3374 if (atomic_read(&fs_info
->balance_running
)) {
3375 atomic_inc(&fs_info
->balance_pause_req
);
3376 mutex_unlock(&fs_info
->balance_mutex
);
3378 wait_event(fs_info
->balance_wait_q
,
3379 atomic_read(&fs_info
->balance_running
) == 0);
3381 mutex_lock(&fs_info
->balance_mutex
);
3382 /* we are good with balance_ctl ripped off from under us */
3383 BUG_ON(atomic_read(&fs_info
->balance_running
));
3384 atomic_dec(&fs_info
->balance_pause_req
);
3389 mutex_unlock(&fs_info
->balance_mutex
);
3393 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3395 mutex_lock(&fs_info
->balance_mutex
);
3396 if (!fs_info
->balance_ctl
) {
3397 mutex_unlock(&fs_info
->balance_mutex
);
3401 atomic_inc(&fs_info
->balance_cancel_req
);
3403 * if we are running just wait and return, balance item is
3404 * deleted in btrfs_balance in this case
3406 if (atomic_read(&fs_info
->balance_running
)) {
3407 mutex_unlock(&fs_info
->balance_mutex
);
3408 wait_event(fs_info
->balance_wait_q
,
3409 atomic_read(&fs_info
->balance_running
) == 0);
3410 mutex_lock(&fs_info
->balance_mutex
);
3412 /* __cancel_balance needs volume_mutex */
3413 mutex_unlock(&fs_info
->balance_mutex
);
3414 mutex_lock(&fs_info
->volume_mutex
);
3415 mutex_lock(&fs_info
->balance_mutex
);
3417 if (fs_info
->balance_ctl
)
3418 __cancel_balance(fs_info
);
3420 mutex_unlock(&fs_info
->volume_mutex
);
3423 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3424 atomic_dec(&fs_info
->balance_cancel_req
);
3425 mutex_unlock(&fs_info
->balance_mutex
);
3430 * shrinking a device means finding all of the device extents past
3431 * the new size, and then following the back refs to the chunks.
3432 * The chunk relocation code actually frees the device extent
3434 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3436 struct btrfs_trans_handle
*trans
;
3437 struct btrfs_root
*root
= device
->dev_root
;
3438 struct btrfs_dev_extent
*dev_extent
= NULL
;
3439 struct btrfs_path
*path
;
3447 bool retried
= false;
3448 struct extent_buffer
*l
;
3449 struct btrfs_key key
;
3450 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3451 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3452 u64 old_size
= device
->total_bytes
;
3453 u64 diff
= device
->total_bytes
- new_size
;
3455 if (device
->is_tgtdev_for_dev_replace
)
3458 path
= btrfs_alloc_path();
3466 device
->total_bytes
= new_size
;
3467 if (device
->writeable
) {
3468 device
->fs_devices
->total_rw_bytes
-= diff
;
3469 spin_lock(&root
->fs_info
->free_chunk_lock
);
3470 root
->fs_info
->free_chunk_space
-= diff
;
3471 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3473 unlock_chunks(root
);
3476 key
.objectid
= device
->devid
;
3477 key
.offset
= (u64
)-1;
3478 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3481 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3485 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3490 btrfs_release_path(path
);
3495 slot
= path
->slots
[0];
3496 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3498 if (key
.objectid
!= device
->devid
) {
3499 btrfs_release_path(path
);
3503 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3504 length
= btrfs_dev_extent_length(l
, dev_extent
);
3506 if (key
.offset
+ length
<= new_size
) {
3507 btrfs_release_path(path
);
3511 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3512 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3513 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3514 btrfs_release_path(path
);
3516 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3518 if (ret
&& ret
!= -ENOSPC
)
3522 } while (key
.offset
-- > 0);
3524 if (failed
&& !retried
) {
3528 } else if (failed
&& retried
) {
3532 device
->total_bytes
= old_size
;
3533 if (device
->writeable
)
3534 device
->fs_devices
->total_rw_bytes
+= diff
;
3535 spin_lock(&root
->fs_info
->free_chunk_lock
);
3536 root
->fs_info
->free_chunk_space
+= diff
;
3537 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3538 unlock_chunks(root
);
3542 /* Shrinking succeeded, else we would be at "done". */
3543 trans
= btrfs_start_transaction(root
, 0);
3544 if (IS_ERR(trans
)) {
3545 ret
= PTR_ERR(trans
);
3551 device
->disk_total_bytes
= new_size
;
3552 /* Now btrfs_update_device() will change the on-disk size. */
3553 ret
= btrfs_update_device(trans
, device
);
3555 unlock_chunks(root
);
3556 btrfs_end_transaction(trans
, root
);
3559 WARN_ON(diff
> old_total
);
3560 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3561 unlock_chunks(root
);
3562 btrfs_end_transaction(trans
, root
);
3564 btrfs_free_path(path
);
3568 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3569 struct btrfs_key
*key
,
3570 struct btrfs_chunk
*chunk
, int item_size
)
3572 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3573 struct btrfs_disk_key disk_key
;
3577 array_size
= btrfs_super_sys_array_size(super_copy
);
3578 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3581 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3582 btrfs_cpu_key_to_disk(&disk_key
, key
);
3583 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3584 ptr
+= sizeof(disk_key
);
3585 memcpy(ptr
, chunk
, item_size
);
3586 item_size
+= sizeof(disk_key
);
3587 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3592 * sort the devices in descending order by max_avail, total_avail
3594 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3596 const struct btrfs_device_info
*di_a
= a
;
3597 const struct btrfs_device_info
*di_b
= b
;
3599 if (di_a
->max_avail
> di_b
->max_avail
)
3601 if (di_a
->max_avail
< di_b
->max_avail
)
3603 if (di_a
->total_avail
> di_b
->total_avail
)
3605 if (di_a
->total_avail
< di_b
->total_avail
)
3610 struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
3611 [BTRFS_RAID_RAID10
] = {
3614 .devs_max
= 0, /* 0 == as many as possible */
3616 .devs_increment
= 2,
3619 [BTRFS_RAID_RAID1
] = {
3624 .devs_increment
= 2,
3627 [BTRFS_RAID_DUP
] = {
3632 .devs_increment
= 1,
3635 [BTRFS_RAID_RAID0
] = {
3640 .devs_increment
= 1,
3643 [BTRFS_RAID_SINGLE
] = {
3648 .devs_increment
= 1,
3651 [BTRFS_RAID_RAID5
] = {
3656 .devs_increment
= 1,
3659 [BTRFS_RAID_RAID6
] = {
3664 .devs_increment
= 1,
3669 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
3671 /* TODO allow them to set a preferred stripe size */
3675 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
3679 if (!(type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)))
3682 features
= btrfs_super_incompat_flags(info
->super_copy
);
3683 if (features
& BTRFS_FEATURE_INCOMPAT_RAID56
)
3686 features
|= BTRFS_FEATURE_INCOMPAT_RAID56
;
3687 btrfs_set_super_incompat_flags(info
->super_copy
, features
);
3688 printk(KERN_INFO
"btrfs: setting RAID5/6 feature flag\n");
3691 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3692 struct btrfs_root
*extent_root
,
3693 struct map_lookup
**map_ret
,
3694 u64
*num_bytes_out
, u64
*stripe_size_out
,
3695 u64 start
, u64 type
)
3697 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3698 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3699 struct list_head
*cur
;
3700 struct map_lookup
*map
= NULL
;
3701 struct extent_map_tree
*em_tree
;
3702 struct extent_map
*em
;
3703 struct btrfs_device_info
*devices_info
= NULL
;
3705 int num_stripes
; /* total number of stripes to allocate */
3706 int data_stripes
; /* number of stripes that count for
3708 int sub_stripes
; /* sub_stripes info for map */
3709 int dev_stripes
; /* stripes per dev */
3710 int devs_max
; /* max devs to use */
3711 int devs_min
; /* min devs needed */
3712 int devs_increment
; /* ndevs has to be a multiple of this */
3713 int ncopies
; /* how many copies to data has */
3715 u64 max_stripe_size
;
3719 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
3725 BUG_ON(!alloc_profile_is_valid(type
, 0));
3727 if (list_empty(&fs_devices
->alloc_list
))
3730 index
= __get_raid_index(type
);
3732 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
3733 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
3734 devs_max
= btrfs_raid_array
[index
].devs_max
;
3735 devs_min
= btrfs_raid_array
[index
].devs_min
;
3736 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
3737 ncopies
= btrfs_raid_array
[index
].ncopies
;
3739 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3740 max_stripe_size
= 1024 * 1024 * 1024;
3741 max_chunk_size
= 10 * max_stripe_size
;
3742 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3743 /* for larger filesystems, use larger metadata chunks */
3744 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3745 max_stripe_size
= 1024 * 1024 * 1024;
3747 max_stripe_size
= 256 * 1024 * 1024;
3748 max_chunk_size
= max_stripe_size
;
3749 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3750 max_stripe_size
= 32 * 1024 * 1024;
3751 max_chunk_size
= 2 * max_stripe_size
;
3753 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3758 /* we don't want a chunk larger than 10% of writeable space */
3759 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3762 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3767 cur
= fs_devices
->alloc_list
.next
;
3770 * in the first pass through the devices list, we gather information
3771 * about the available holes on each device.
3774 while (cur
!= &fs_devices
->alloc_list
) {
3775 struct btrfs_device
*device
;
3779 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3783 if (!device
->writeable
) {
3785 "btrfs: read-only device in alloc_list\n");
3789 if (!device
->in_fs_metadata
||
3790 device
->is_tgtdev_for_dev_replace
)
3793 if (device
->total_bytes
> device
->bytes_used
)
3794 total_avail
= device
->total_bytes
- device
->bytes_used
;
3798 /* If there is no space on this device, skip it. */
3799 if (total_avail
== 0)
3802 ret
= find_free_dev_extent(device
,
3803 max_stripe_size
* dev_stripes
,
3804 &dev_offset
, &max_avail
);
3805 if (ret
&& ret
!= -ENOSPC
)
3809 max_avail
= max_stripe_size
* dev_stripes
;
3811 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3814 if (ndevs
== fs_devices
->rw_devices
) {
3815 WARN(1, "%s: found more than %llu devices\n",
3816 __func__
, fs_devices
->rw_devices
);
3819 devices_info
[ndevs
].dev_offset
= dev_offset
;
3820 devices_info
[ndevs
].max_avail
= max_avail
;
3821 devices_info
[ndevs
].total_avail
= total_avail
;
3822 devices_info
[ndevs
].dev
= device
;
3827 * now sort the devices by hole size / available space
3829 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3830 btrfs_cmp_device_info
, NULL
);
3832 /* round down to number of usable stripes */
3833 ndevs
-= ndevs
% devs_increment
;
3835 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3840 if (devs_max
&& ndevs
> devs_max
)
3843 * the primary goal is to maximize the number of stripes, so use as many
3844 * devices as possible, even if the stripes are not maximum sized.
3846 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3847 num_stripes
= ndevs
* dev_stripes
;
3850 * this will have to be fixed for RAID1 and RAID10 over
3853 data_stripes
= num_stripes
/ ncopies
;
3855 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
3856 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
3857 btrfs_super_stripesize(info
->super_copy
));
3858 data_stripes
= num_stripes
- 1;
3860 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
3861 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
3862 btrfs_super_stripesize(info
->super_copy
));
3863 data_stripes
= num_stripes
- 2;
3867 * Use the number of data stripes to figure out how big this chunk
3868 * is really going to be in terms of logical address space,
3869 * and compare that answer with the max chunk size
3871 if (stripe_size
* data_stripes
> max_chunk_size
) {
3872 u64 mask
= (1ULL << 24) - 1;
3873 stripe_size
= max_chunk_size
;
3874 do_div(stripe_size
, data_stripes
);
3876 /* bump the answer up to a 16MB boundary */
3877 stripe_size
= (stripe_size
+ mask
) & ~mask
;
3879 /* but don't go higher than the limits we found
3880 * while searching for free extents
3882 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
3883 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3886 do_div(stripe_size
, dev_stripes
);
3888 /* align to BTRFS_STRIPE_LEN */
3889 do_div(stripe_size
, raid_stripe_len
);
3890 stripe_size
*= raid_stripe_len
;
3892 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3897 map
->num_stripes
= num_stripes
;
3899 for (i
= 0; i
< ndevs
; ++i
) {
3900 for (j
= 0; j
< dev_stripes
; ++j
) {
3901 int s
= i
* dev_stripes
+ j
;
3902 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3903 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3907 map
->sector_size
= extent_root
->sectorsize
;
3908 map
->stripe_len
= raid_stripe_len
;
3909 map
->io_align
= raid_stripe_len
;
3910 map
->io_width
= raid_stripe_len
;
3912 map
->sub_stripes
= sub_stripes
;
3915 num_bytes
= stripe_size
* data_stripes
;
3917 *stripe_size_out
= stripe_size
;
3918 *num_bytes_out
= num_bytes
;
3920 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3922 em
= alloc_extent_map();
3927 em
->bdev
= (struct block_device
*)map
;
3929 em
->len
= num_bytes
;
3930 em
->block_start
= 0;
3931 em
->block_len
= em
->len
;
3933 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3934 write_lock(&em_tree
->lock
);
3935 ret
= add_extent_mapping(em_tree
, em
);
3936 write_unlock(&em_tree
->lock
);
3938 free_extent_map(em
);
3942 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3943 struct btrfs_device
*device
;
3946 device
= map
->stripes
[i
].dev
;
3947 dev_offset
= map
->stripes
[i
].physical
;
3949 ret
= btrfs_alloc_dev_extent(trans
, device
,
3950 info
->chunk_root
->root_key
.objectid
,
3951 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3952 start
, dev_offset
, stripe_size
);
3954 goto error_dev_extent
;
3957 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3958 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3961 i
= map
->num_stripes
- 1;
3962 goto error_dev_extent
;
3965 free_extent_map(em
);
3966 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
3968 kfree(devices_info
);
3972 for (; i
>= 0; i
--) {
3973 struct btrfs_device
*device
;
3976 device
= map
->stripes
[i
].dev
;
3977 err
= btrfs_free_dev_extent(trans
, device
, start
);
3979 btrfs_abort_transaction(trans
, extent_root
, err
);
3983 write_lock(&em_tree
->lock
);
3984 remove_extent_mapping(em_tree
, em
);
3985 write_unlock(&em_tree
->lock
);
3987 /* One for our allocation */
3988 free_extent_map(em
);
3989 /* One for the tree reference */
3990 free_extent_map(em
);
3993 kfree(devices_info
);
3997 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3998 struct btrfs_root
*extent_root
,
3999 struct map_lookup
*map
, u64 chunk_offset
,
4000 u64 chunk_size
, u64 stripe_size
)
4003 struct btrfs_key key
;
4004 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4005 struct btrfs_device
*device
;
4006 struct btrfs_chunk
*chunk
;
4007 struct btrfs_stripe
*stripe
;
4008 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4012 chunk
= kzalloc(item_size
, GFP_NOFS
);
4017 while (index
< map
->num_stripes
) {
4018 device
= map
->stripes
[index
].dev
;
4019 device
->bytes_used
+= stripe_size
;
4020 ret
= btrfs_update_device(trans
, device
);
4026 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4027 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4029 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4032 stripe
= &chunk
->stripe
;
4033 while (index
< map
->num_stripes
) {
4034 device
= map
->stripes
[index
].dev
;
4035 dev_offset
= map
->stripes
[index
].physical
;
4037 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4038 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4039 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4044 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4045 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4046 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4047 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4048 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4049 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4050 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4051 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4052 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4054 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4055 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4056 key
.offset
= chunk_offset
;
4058 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4060 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4062 * TODO: Cleanup of inserted chunk root in case of
4065 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4075 * Chunk allocation falls into two parts. The first part does works
4076 * that make the new allocated chunk useable, but not do any operation
4077 * that modifies the chunk tree. The second part does the works that
4078 * require modifying the chunk tree. This division is important for the
4079 * bootstrap process of adding storage to a seed btrfs.
4081 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4082 struct btrfs_root
*extent_root
, u64 type
)
4087 struct map_lookup
*map
;
4088 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4091 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4096 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
4097 &stripe_size
, chunk_offset
, type
);
4101 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
4102 chunk_size
, stripe_size
);
4108 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4109 struct btrfs_root
*root
,
4110 struct btrfs_device
*device
)
4113 u64 sys_chunk_offset
;
4117 u64 sys_stripe_size
;
4119 struct map_lookup
*map
;
4120 struct map_lookup
*sys_map
;
4121 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4122 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4125 ret
= find_next_chunk(fs_info
->chunk_root
,
4126 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
4130 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4131 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
4132 &stripe_size
, chunk_offset
, alloc_profile
);
4136 sys_chunk_offset
= chunk_offset
+ chunk_size
;
4138 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4139 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
4140 &sys_chunk_size
, &sys_stripe_size
,
4141 sys_chunk_offset
, alloc_profile
);
4143 btrfs_abort_transaction(trans
, root
, ret
);
4147 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
4149 btrfs_abort_transaction(trans
, root
, ret
);
4154 * Modifying chunk tree needs allocating new blocks from both
4155 * system block group and metadata block group. So we only can
4156 * do operations require modifying the chunk tree after both
4157 * block groups were created.
4159 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
4160 chunk_size
, stripe_size
);
4162 btrfs_abort_transaction(trans
, root
, ret
);
4166 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
4167 sys_chunk_offset
, sys_chunk_size
,
4170 btrfs_abort_transaction(trans
, root
, ret
);
4177 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4179 struct extent_map
*em
;
4180 struct map_lookup
*map
;
4181 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4185 read_lock(&map_tree
->map_tree
.lock
);
4186 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4187 read_unlock(&map_tree
->map_tree
.lock
);
4191 if (btrfs_test_opt(root
, DEGRADED
)) {
4192 free_extent_map(em
);
4196 map
= (struct map_lookup
*)em
->bdev
;
4197 for (i
= 0; i
< map
->num_stripes
; i
++) {
4198 if (!map
->stripes
[i
].dev
->writeable
) {
4203 free_extent_map(em
);
4207 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4209 extent_map_tree_init(&tree
->map_tree
);
4212 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4214 struct extent_map
*em
;
4217 write_lock(&tree
->map_tree
.lock
);
4218 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4220 remove_extent_mapping(&tree
->map_tree
, em
);
4221 write_unlock(&tree
->map_tree
.lock
);
4226 free_extent_map(em
);
4227 /* once for the tree */
4228 free_extent_map(em
);
4232 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4234 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4235 struct extent_map
*em
;
4236 struct map_lookup
*map
;
4237 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4240 read_lock(&em_tree
->lock
);
4241 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4242 read_unlock(&em_tree
->lock
);
4245 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4246 map
= (struct map_lookup
*)em
->bdev
;
4247 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4248 ret
= map
->num_stripes
;
4249 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4250 ret
= map
->sub_stripes
;
4251 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4253 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4257 free_extent_map(em
);
4259 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4260 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4262 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4267 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4268 struct btrfs_mapping_tree
*map_tree
,
4271 struct extent_map
*em
;
4272 struct map_lookup
*map
;
4273 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4274 unsigned long len
= root
->sectorsize
;
4276 read_lock(&em_tree
->lock
);
4277 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4278 read_unlock(&em_tree
->lock
);
4281 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4282 map
= (struct map_lookup
*)em
->bdev
;
4283 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4284 BTRFS_BLOCK_GROUP_RAID6
)) {
4285 len
= map
->stripe_len
* nr_data_stripes(map
);
4287 free_extent_map(em
);
4291 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4292 u64 logical
, u64 len
, int mirror_num
)
4294 struct extent_map
*em
;
4295 struct map_lookup
*map
;
4296 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4299 read_lock(&em_tree
->lock
);
4300 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4301 read_unlock(&em_tree
->lock
);
4304 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4305 map
= (struct map_lookup
*)em
->bdev
;
4306 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4307 BTRFS_BLOCK_GROUP_RAID6
))
4309 free_extent_map(em
);
4313 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4314 struct map_lookup
*map
, int first
, int num
,
4315 int optimal
, int dev_replace_is_ongoing
)
4319 struct btrfs_device
*srcdev
;
4321 if (dev_replace_is_ongoing
&&
4322 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4323 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4324 srcdev
= fs_info
->dev_replace
.srcdev
;
4329 * try to avoid the drive that is the source drive for a
4330 * dev-replace procedure, only choose it if no other non-missing
4331 * mirror is available
4333 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4334 if (map
->stripes
[optimal
].dev
->bdev
&&
4335 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4337 for (i
= first
; i
< first
+ num
; i
++) {
4338 if (map
->stripes
[i
].dev
->bdev
&&
4339 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4344 /* we couldn't find one that doesn't fail. Just return something
4345 * and the io error handling code will clean up eventually
4350 static inline int parity_smaller(u64 a
, u64 b
)
4355 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4356 static void sort_parity_stripes(struct btrfs_bio
*bbio
, u64
*raid_map
)
4358 struct btrfs_bio_stripe s
;
4365 for (i
= 0; i
< bbio
->num_stripes
- 1; i
++) {
4366 if (parity_smaller(raid_map
[i
], raid_map
[i
+1])) {
4367 s
= bbio
->stripes
[i
];
4369 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4370 raid_map
[i
] = raid_map
[i
+1];
4371 bbio
->stripes
[i
+1] = s
;
4379 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4380 u64 logical
, u64
*length
,
4381 struct btrfs_bio
**bbio_ret
,
4382 int mirror_num
, u64
**raid_map_ret
)
4384 struct extent_map
*em
;
4385 struct map_lookup
*map
;
4386 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4387 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4390 u64 stripe_end_offset
;
4395 u64
*raid_map
= NULL
;
4401 struct btrfs_bio
*bbio
= NULL
;
4402 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4403 int dev_replace_is_ongoing
= 0;
4404 int num_alloc_stripes
;
4405 int patch_the_first_stripe_for_dev_replace
= 0;
4406 u64 physical_to_patch_in_first_stripe
= 0;
4407 u64 raid56_full_stripe_start
= (u64
)-1;
4409 read_lock(&em_tree
->lock
);
4410 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4411 read_unlock(&em_tree
->lock
);
4414 printk(KERN_CRIT
"btrfs: unable to find logical %llu len %llu\n",
4415 (unsigned long long)logical
,
4416 (unsigned long long)*length
);
4420 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4421 map
= (struct map_lookup
*)em
->bdev
;
4422 offset
= logical
- em
->start
;
4424 if (mirror_num
> map
->num_stripes
)
4427 stripe_len
= map
->stripe_len
;
4430 * stripe_nr counts the total number of stripes we have to stride
4431 * to get to this block
4433 do_div(stripe_nr
, stripe_len
);
4435 stripe_offset
= stripe_nr
* stripe_len
;
4436 BUG_ON(offset
< stripe_offset
);
4438 /* stripe_offset is the offset of this block in its stripe*/
4439 stripe_offset
= offset
- stripe_offset
;
4441 /* if we're here for raid56, we need to know the stripe aligned start */
4442 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4443 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4444 raid56_full_stripe_start
= offset
;
4446 /* allow a write of a full stripe, but make sure we don't
4447 * allow straddling of stripes
4449 do_div(raid56_full_stripe_start
, full_stripe_len
);
4450 raid56_full_stripe_start
*= full_stripe_len
;
4453 if (rw
& REQ_DISCARD
) {
4454 /* we don't discard raid56 yet */
4456 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4460 *length
= min_t(u64
, em
->len
- offset
, *length
);
4461 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4463 /* For writes to RAID[56], allow a full stripeset across all disks.
4464 For other RAID types and for RAID[56] reads, just allow a single
4465 stripe (on a single disk). */
4466 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
) &&
4468 max_len
= stripe_len
* nr_data_stripes(map
) -
4469 (offset
- raid56_full_stripe_start
);
4471 /* we limit the length of each bio to what fits in a stripe */
4472 max_len
= stripe_len
- stripe_offset
;
4474 *length
= min_t(u64
, em
->len
- offset
, max_len
);
4476 *length
= em
->len
- offset
;
4479 /* This is for when we're called from btrfs_merge_bio_hook() and all
4480 it cares about is the length */
4484 btrfs_dev_replace_lock(dev_replace
);
4485 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
4486 if (!dev_replace_is_ongoing
)
4487 btrfs_dev_replace_unlock(dev_replace
);
4489 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
4490 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
4491 dev_replace
->tgtdev
!= NULL
) {
4493 * in dev-replace case, for repair case (that's the only
4494 * case where the mirror is selected explicitly when
4495 * calling btrfs_map_block), blocks left of the left cursor
4496 * can also be read from the target drive.
4497 * For REQ_GET_READ_MIRRORS, the target drive is added as
4498 * the last one to the array of stripes. For READ, it also
4499 * needs to be supported using the same mirror number.
4500 * If the requested block is not left of the left cursor,
4501 * EIO is returned. This can happen because btrfs_num_copies()
4502 * returns one more in the dev-replace case.
4504 u64 tmp_length
= *length
;
4505 struct btrfs_bio
*tmp_bbio
= NULL
;
4506 int tmp_num_stripes
;
4507 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4508 int index_srcdev
= 0;
4510 u64 physical_of_found
= 0;
4512 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
4513 logical
, &tmp_length
, &tmp_bbio
, 0, NULL
);
4515 WARN_ON(tmp_bbio
!= NULL
);
4519 tmp_num_stripes
= tmp_bbio
->num_stripes
;
4520 if (mirror_num
> tmp_num_stripes
) {
4522 * REQ_GET_READ_MIRRORS does not contain this
4523 * mirror, that means that the requested area
4524 * is not left of the left cursor
4532 * process the rest of the function using the mirror_num
4533 * of the source drive. Therefore look it up first.
4534 * At the end, patch the device pointer to the one of the
4537 for (i
= 0; i
< tmp_num_stripes
; i
++) {
4538 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4540 * In case of DUP, in order to keep it
4541 * simple, only add the mirror with the
4542 * lowest physical address
4545 physical_of_found
<=
4546 tmp_bbio
->stripes
[i
].physical
)
4551 tmp_bbio
->stripes
[i
].physical
;
4556 mirror_num
= index_srcdev
+ 1;
4557 patch_the_first_stripe_for_dev_replace
= 1;
4558 physical_to_patch_in_first_stripe
= physical_of_found
;
4567 } else if (mirror_num
> map
->num_stripes
) {
4573 stripe_nr_orig
= stripe_nr
;
4574 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
4575 do_div(stripe_nr_end
, map
->stripe_len
);
4576 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
4579 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4580 if (rw
& REQ_DISCARD
)
4581 num_stripes
= min_t(u64
, map
->num_stripes
,
4582 stripe_nr_end
- stripe_nr_orig
);
4583 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4584 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
4585 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
4586 num_stripes
= map
->num_stripes
;
4587 else if (mirror_num
)
4588 stripe_index
= mirror_num
- 1;
4590 stripe_index
= find_live_mirror(fs_info
, map
, 0,
4592 current
->pid
% map
->num_stripes
,
4593 dev_replace_is_ongoing
);
4594 mirror_num
= stripe_index
+ 1;
4597 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
4598 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
4599 num_stripes
= map
->num_stripes
;
4600 } else if (mirror_num
) {
4601 stripe_index
= mirror_num
- 1;
4606 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4607 int factor
= map
->num_stripes
/ map
->sub_stripes
;
4609 stripe_index
= do_div(stripe_nr
, factor
);
4610 stripe_index
*= map
->sub_stripes
;
4612 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
4613 num_stripes
= map
->sub_stripes
;
4614 else if (rw
& REQ_DISCARD
)
4615 num_stripes
= min_t(u64
, map
->sub_stripes
*
4616 (stripe_nr_end
- stripe_nr_orig
),
4618 else if (mirror_num
)
4619 stripe_index
+= mirror_num
- 1;
4621 int old_stripe_index
= stripe_index
;
4622 stripe_index
= find_live_mirror(fs_info
, map
,
4624 map
->sub_stripes
, stripe_index
+
4625 current
->pid
% map
->sub_stripes
,
4626 dev_replace_is_ongoing
);
4627 mirror_num
= stripe_index
- old_stripe_index
+ 1;
4630 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4631 BTRFS_BLOCK_GROUP_RAID6
)) {
4634 if (bbio_ret
&& ((rw
& REQ_WRITE
) || mirror_num
> 1)
4638 /* push stripe_nr back to the start of the full stripe */
4639 stripe_nr
= raid56_full_stripe_start
;
4640 do_div(stripe_nr
, stripe_len
);
4642 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4644 /* RAID[56] write or recovery. Return all stripes */
4645 num_stripes
= map
->num_stripes
;
4646 max_errors
= nr_parity_stripes(map
);
4648 raid_map
= kmalloc(sizeof(u64
) * num_stripes
,
4655 /* Work out the disk rotation on this stripe-set */
4657 rot
= do_div(tmp
, num_stripes
);
4659 /* Fill in the logical address of each stripe */
4660 tmp
= stripe_nr
* nr_data_stripes(map
);
4661 for (i
= 0; i
< nr_data_stripes(map
); i
++)
4662 raid_map
[(i
+rot
) % num_stripes
] =
4663 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
4665 raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
4666 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4667 raid_map
[(i
+rot
+1) % num_stripes
] =
4670 *length
= map
->stripe_len
;
4675 * Mirror #0 or #1 means the original data block.
4676 * Mirror #2 is RAID5 parity block.
4677 * Mirror #3 is RAID6 Q block.
4679 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4681 stripe_index
= nr_data_stripes(map
) +
4684 /* We distribute the parity blocks across stripes */
4685 tmp
= stripe_nr
+ stripe_index
;
4686 stripe_index
= do_div(tmp
, map
->num_stripes
);
4690 * after this do_div call, stripe_nr is the number of stripes
4691 * on this device we have to walk to find the data, and
4692 * stripe_index is the number of our device in the stripe array
4694 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4695 mirror_num
= stripe_index
+ 1;
4697 BUG_ON(stripe_index
>= map
->num_stripes
);
4699 num_alloc_stripes
= num_stripes
;
4700 if (dev_replace_is_ongoing
) {
4701 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
4702 num_alloc_stripes
<<= 1;
4703 if (rw
& REQ_GET_READ_MIRRORS
)
4704 num_alloc_stripes
++;
4706 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
4711 atomic_set(&bbio
->error
, 0);
4713 if (rw
& REQ_DISCARD
) {
4715 int sub_stripes
= 0;
4716 u64 stripes_per_dev
= 0;
4717 u32 remaining_stripes
= 0;
4718 u32 last_stripe
= 0;
4721 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
4722 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4725 sub_stripes
= map
->sub_stripes
;
4727 factor
= map
->num_stripes
/ sub_stripes
;
4728 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
4731 &remaining_stripes
);
4732 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
4733 last_stripe
*= sub_stripes
;
4736 for (i
= 0; i
< num_stripes
; i
++) {
4737 bbio
->stripes
[i
].physical
=
4738 map
->stripes
[stripe_index
].physical
+
4739 stripe_offset
+ stripe_nr
* map
->stripe_len
;
4740 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
4742 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
4743 BTRFS_BLOCK_GROUP_RAID10
)) {
4744 bbio
->stripes
[i
].length
= stripes_per_dev
*
4747 if (i
/ sub_stripes
< remaining_stripes
)
4748 bbio
->stripes
[i
].length
+=
4752 * Special for the first stripe and
4755 * |-------|...|-------|
4759 if (i
< sub_stripes
)
4760 bbio
->stripes
[i
].length
-=
4763 if (stripe_index
>= last_stripe
&&
4764 stripe_index
<= (last_stripe
+
4766 bbio
->stripes
[i
].length
-=
4769 if (i
== sub_stripes
- 1)
4772 bbio
->stripes
[i
].length
= *length
;
4775 if (stripe_index
== map
->num_stripes
) {
4776 /* This could only happen for RAID0/10 */
4782 for (i
= 0; i
< num_stripes
; i
++) {
4783 bbio
->stripes
[i
].physical
=
4784 map
->stripes
[stripe_index
].physical
+
4786 stripe_nr
* map
->stripe_len
;
4787 bbio
->stripes
[i
].dev
=
4788 map
->stripes
[stripe_index
].dev
;
4793 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) {
4794 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4795 BTRFS_BLOCK_GROUP_RAID10
|
4796 BTRFS_BLOCK_GROUP_RAID5
|
4797 BTRFS_BLOCK_GROUP_DUP
)) {
4799 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4804 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
4805 dev_replace
->tgtdev
!= NULL
) {
4806 int index_where_to_add
;
4807 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4810 * duplicate the write operations while the dev replace
4811 * procedure is running. Since the copying of the old disk
4812 * to the new disk takes place at run time while the
4813 * filesystem is mounted writable, the regular write
4814 * operations to the old disk have to be duplicated to go
4815 * to the new disk as well.
4816 * Note that device->missing is handled by the caller, and
4817 * that the write to the old disk is already set up in the
4820 index_where_to_add
= num_stripes
;
4821 for (i
= 0; i
< num_stripes
; i
++) {
4822 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4823 /* write to new disk, too */
4824 struct btrfs_bio_stripe
*new =
4825 bbio
->stripes
+ index_where_to_add
;
4826 struct btrfs_bio_stripe
*old
=
4829 new->physical
= old
->physical
;
4830 new->length
= old
->length
;
4831 new->dev
= dev_replace
->tgtdev
;
4832 index_where_to_add
++;
4836 num_stripes
= index_where_to_add
;
4837 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
4838 dev_replace
->tgtdev
!= NULL
) {
4839 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4840 int index_srcdev
= 0;
4842 u64 physical_of_found
= 0;
4845 * During the dev-replace procedure, the target drive can
4846 * also be used to read data in case it is needed to repair
4847 * a corrupt block elsewhere. This is possible if the
4848 * requested area is left of the left cursor. In this area,
4849 * the target drive is a full copy of the source drive.
4851 for (i
= 0; i
< num_stripes
; i
++) {
4852 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4854 * In case of DUP, in order to keep it
4855 * simple, only add the mirror with the
4856 * lowest physical address
4859 physical_of_found
<=
4860 bbio
->stripes
[i
].physical
)
4864 physical_of_found
= bbio
->stripes
[i
].physical
;
4868 u64 length
= map
->stripe_len
;
4870 if (physical_of_found
+ length
<=
4871 dev_replace
->cursor_left
) {
4872 struct btrfs_bio_stripe
*tgtdev_stripe
=
4873 bbio
->stripes
+ num_stripes
;
4875 tgtdev_stripe
->physical
= physical_of_found
;
4876 tgtdev_stripe
->length
=
4877 bbio
->stripes
[index_srcdev
].length
;
4878 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
4886 bbio
->num_stripes
= num_stripes
;
4887 bbio
->max_errors
= max_errors
;
4888 bbio
->mirror_num
= mirror_num
;
4891 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4892 * mirror_num == num_stripes + 1 && dev_replace target drive is
4893 * available as a mirror
4895 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
4896 WARN_ON(num_stripes
> 1);
4897 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
4898 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
4899 bbio
->mirror_num
= map
->num_stripes
+ 1;
4902 sort_parity_stripes(bbio
, raid_map
);
4903 *raid_map_ret
= raid_map
;
4906 if (dev_replace_is_ongoing
)
4907 btrfs_dev_replace_unlock(dev_replace
);
4908 free_extent_map(em
);
4912 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4913 u64 logical
, u64
*length
,
4914 struct btrfs_bio
**bbio_ret
, int mirror_num
)
4916 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
4920 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
4921 u64 chunk_start
, u64 physical
, u64 devid
,
4922 u64
**logical
, int *naddrs
, int *stripe_len
)
4924 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4925 struct extent_map
*em
;
4926 struct map_lookup
*map
;
4934 read_lock(&em_tree
->lock
);
4935 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
4936 read_unlock(&em_tree
->lock
);
4939 printk(KERN_ERR
"btrfs: couldn't find em for chunk %Lu\n",
4944 if (em
->start
!= chunk_start
) {
4945 printk(KERN_ERR
"btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
4946 em
->start
, chunk_start
);
4947 free_extent_map(em
);
4950 map
= (struct map_lookup
*)em
->bdev
;
4953 rmap_len
= map
->stripe_len
;
4955 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4956 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
4957 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4958 do_div(length
, map
->num_stripes
);
4959 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4960 BTRFS_BLOCK_GROUP_RAID6
)) {
4961 do_div(length
, nr_data_stripes(map
));
4962 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
4965 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
4966 BUG_ON(!buf
); /* -ENOMEM */
4968 for (i
= 0; i
< map
->num_stripes
; i
++) {
4969 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
4971 if (map
->stripes
[i
].physical
> physical
||
4972 map
->stripes
[i
].physical
+ length
<= physical
)
4975 stripe_nr
= physical
- map
->stripes
[i
].physical
;
4976 do_div(stripe_nr
, map
->stripe_len
);
4978 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4979 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4980 do_div(stripe_nr
, map
->sub_stripes
);
4981 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4982 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4983 } /* else if RAID[56], multiply by nr_data_stripes().
4984 * Alternatively, just use rmap_len below instead of
4985 * map->stripe_len */
4987 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
4988 WARN_ON(nr
>= map
->num_stripes
);
4989 for (j
= 0; j
< nr
; j
++) {
4990 if (buf
[j
] == bytenr
)
4994 WARN_ON(nr
>= map
->num_stripes
);
5001 *stripe_len
= rmap_len
;
5003 free_extent_map(em
);
5007 static void *merge_stripe_index_into_bio_private(void *bi_private
,
5008 unsigned int stripe_index
)
5011 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
5013 * The alternative solution (instead of stealing bits from the
5014 * pointer) would be to allocate an intermediate structure
5015 * that contains the old private pointer plus the stripe_index.
5017 BUG_ON((((uintptr_t)bi_private
) & 3) != 0);
5018 BUG_ON(stripe_index
> 3);
5019 return (void *)(((uintptr_t)bi_private
) | stripe_index
);
5022 static struct btrfs_bio
*extract_bbio_from_bio_private(void *bi_private
)
5024 return (struct btrfs_bio
*)(((uintptr_t)bi_private
) & ~((uintptr_t)3));
5027 static unsigned int extract_stripe_index_from_bio_private(void *bi_private
)
5029 return (unsigned int)((uintptr_t)bi_private
) & 3;
5032 static void btrfs_end_bio(struct bio
*bio
, int err
)
5034 struct btrfs_bio
*bbio
= extract_bbio_from_bio_private(bio
->bi_private
);
5035 int is_orig_bio
= 0;
5038 atomic_inc(&bbio
->error
);
5039 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5040 unsigned int stripe_index
=
5041 extract_stripe_index_from_bio_private(
5043 struct btrfs_device
*dev
;
5045 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5046 dev
= bbio
->stripes
[stripe_index
].dev
;
5048 if (bio
->bi_rw
& WRITE
)
5049 btrfs_dev_stat_inc(dev
,
5050 BTRFS_DEV_STAT_WRITE_ERRS
);
5052 btrfs_dev_stat_inc(dev
,
5053 BTRFS_DEV_STAT_READ_ERRS
);
5054 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5055 btrfs_dev_stat_inc(dev
,
5056 BTRFS_DEV_STAT_FLUSH_ERRS
);
5057 btrfs_dev_stat_print_on_error(dev
);
5062 if (bio
== bbio
->orig_bio
)
5065 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5068 bio
= bbio
->orig_bio
;
5070 bio
->bi_private
= bbio
->private;
5071 bio
->bi_end_io
= bbio
->end_io
;
5072 bio
->bi_bdev
= (struct block_device
*)
5073 (unsigned long)bbio
->mirror_num
;
5074 /* only send an error to the higher layers if it is
5075 * beyond the tolerance of the btrfs bio
5077 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5081 * this bio is actually up to date, we didn't
5082 * go over the max number of errors
5084 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5089 bio_endio(bio
, err
);
5090 } else if (!is_orig_bio
) {
5095 struct async_sched
{
5098 struct btrfs_fs_info
*info
;
5099 struct btrfs_work work
;
5103 * see run_scheduled_bios for a description of why bios are collected for
5106 * This will add one bio to the pending list for a device and make sure
5107 * the work struct is scheduled.
5109 noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5110 struct btrfs_device
*device
,
5111 int rw
, struct bio
*bio
)
5113 int should_queue
= 1;
5114 struct btrfs_pending_bios
*pending_bios
;
5116 if (device
->missing
|| !device
->bdev
) {
5117 bio_endio(bio
, -EIO
);
5121 /* don't bother with additional async steps for reads, right now */
5122 if (!(rw
& REQ_WRITE
)) {
5124 btrfsic_submit_bio(rw
, bio
);
5130 * nr_async_bios allows us to reliably return congestion to the
5131 * higher layers. Otherwise, the async bio makes it appear we have
5132 * made progress against dirty pages when we've really just put it
5133 * on a queue for later
5135 atomic_inc(&root
->fs_info
->nr_async_bios
);
5136 WARN_ON(bio
->bi_next
);
5137 bio
->bi_next
= NULL
;
5140 spin_lock(&device
->io_lock
);
5141 if (bio
->bi_rw
& REQ_SYNC
)
5142 pending_bios
= &device
->pending_sync_bios
;
5144 pending_bios
= &device
->pending_bios
;
5146 if (pending_bios
->tail
)
5147 pending_bios
->tail
->bi_next
= bio
;
5149 pending_bios
->tail
= bio
;
5150 if (!pending_bios
->head
)
5151 pending_bios
->head
= bio
;
5152 if (device
->running_pending
)
5155 spin_unlock(&device
->io_lock
);
5158 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
5162 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5165 struct bio_vec
*prev
;
5166 struct request_queue
*q
= bdev_get_queue(bdev
);
5167 unsigned short max_sectors
= queue_max_sectors(q
);
5168 struct bvec_merge_data bvm
= {
5170 .bi_sector
= sector
,
5171 .bi_rw
= bio
->bi_rw
,
5174 if (bio
->bi_vcnt
== 0) {
5179 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5180 if ((bio
->bi_size
>> 9) > max_sectors
)
5183 if (!q
->merge_bvec_fn
)
5186 bvm
.bi_size
= bio
->bi_size
- prev
->bv_len
;
5187 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5192 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5193 struct bio
*bio
, u64 physical
, int dev_nr
,
5196 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5198 bio
->bi_private
= bbio
;
5199 bio
->bi_private
= merge_stripe_index_into_bio_private(
5200 bio
->bi_private
, (unsigned int)dev_nr
);
5201 bio
->bi_end_io
= btrfs_end_bio
;
5202 bio
->bi_sector
= physical
>> 9;
5205 struct rcu_string
*name
;
5208 name
= rcu_dereference(dev
->name
);
5209 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5210 "(%s id %llu), size=%u\n", rw
,
5211 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5212 name
->str
, dev
->devid
, bio
->bi_size
);
5216 bio
->bi_bdev
= dev
->bdev
;
5218 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5220 btrfsic_submit_bio(rw
, bio
);
5223 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5224 struct bio
*first_bio
, struct btrfs_device
*dev
,
5225 int dev_nr
, int rw
, int async
)
5227 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5229 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5230 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5233 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5237 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5238 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5239 bvec
->bv_offset
) < bvec
->bv_len
) {
5240 u64 len
= bio
->bi_size
;
5242 atomic_inc(&bbio
->stripes_pending
);
5243 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5251 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5255 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5257 atomic_inc(&bbio
->error
);
5258 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5259 bio
->bi_private
= bbio
->private;
5260 bio
->bi_end_io
= bbio
->end_io
;
5261 bio
->bi_bdev
= (struct block_device
*)
5262 (unsigned long)bbio
->mirror_num
;
5263 bio
->bi_sector
= logical
>> 9;
5265 bio_endio(bio
, -EIO
);
5269 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5270 int mirror_num
, int async_submit
)
5272 struct btrfs_device
*dev
;
5273 struct bio
*first_bio
= bio
;
5274 u64 logical
= (u64
)bio
->bi_sector
<< 9;
5277 u64
*raid_map
= NULL
;
5281 struct btrfs_bio
*bbio
= NULL
;
5283 length
= bio
->bi_size
;
5284 map_length
= length
;
5286 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5287 mirror_num
, &raid_map
);
5288 if (ret
) /* -ENOMEM */
5291 total_devs
= bbio
->num_stripes
;
5292 bbio
->orig_bio
= first_bio
;
5293 bbio
->private = first_bio
->bi_private
;
5294 bbio
->end_io
= first_bio
->bi_end_io
;
5295 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5298 /* In this case, map_length has been set to the length of
5299 a single stripe; not the whole write */
5301 return raid56_parity_write(root
, bio
, bbio
,
5302 raid_map
, map_length
);
5304 return raid56_parity_recover(root
, bio
, bbio
,
5305 raid_map
, map_length
,
5310 if (map_length
< length
) {
5311 printk(KERN_CRIT
"btrfs: mapping failed logical %llu bio len %llu "
5312 "len %llu\n", (unsigned long long)logical
,
5313 (unsigned long long)length
,
5314 (unsigned long long)map_length
);
5318 while (dev_nr
< total_devs
) {
5319 dev
= bbio
->stripes
[dev_nr
].dev
;
5320 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5321 bbio_error(bbio
, first_bio
, logical
);
5327 * Check and see if we're ok with this bio based on it's size
5328 * and offset with the given device.
5330 if (!bio_size_ok(dev
->bdev
, first_bio
,
5331 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5332 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5333 dev_nr
, rw
, async_submit
);
5339 if (dev_nr
< total_devs
- 1) {
5340 bio
= bio_clone(first_bio
, GFP_NOFS
);
5341 BUG_ON(!bio
); /* -ENOMEM */
5346 submit_stripe_bio(root
, bbio
, bio
,
5347 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5354 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5357 struct btrfs_device
*device
;
5358 struct btrfs_fs_devices
*cur_devices
;
5360 cur_devices
= fs_info
->fs_devices
;
5361 while (cur_devices
) {
5363 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5364 device
= __find_device(&cur_devices
->devices
,
5369 cur_devices
= cur_devices
->seed
;
5374 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5375 u64 devid
, u8
*dev_uuid
)
5377 struct btrfs_device
*device
;
5378 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5380 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
5383 list_add(&device
->dev_list
,
5384 &fs_devices
->devices
);
5385 device
->dev_root
= root
->fs_info
->dev_root
;
5386 device
->devid
= devid
;
5387 device
->work
.func
= pending_bios_fn
;
5388 device
->fs_devices
= fs_devices
;
5389 device
->missing
= 1;
5390 fs_devices
->num_devices
++;
5391 fs_devices
->missing_devices
++;
5392 spin_lock_init(&device
->io_lock
);
5393 INIT_LIST_HEAD(&device
->dev_alloc_list
);
5394 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
5398 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5399 struct extent_buffer
*leaf
,
5400 struct btrfs_chunk
*chunk
)
5402 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5403 struct map_lookup
*map
;
5404 struct extent_map
*em
;
5408 u8 uuid
[BTRFS_UUID_SIZE
];
5413 logical
= key
->offset
;
5414 length
= btrfs_chunk_length(leaf
, chunk
);
5416 read_lock(&map_tree
->map_tree
.lock
);
5417 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5418 read_unlock(&map_tree
->map_tree
.lock
);
5420 /* already mapped? */
5421 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5422 free_extent_map(em
);
5425 free_extent_map(em
);
5428 em
= alloc_extent_map();
5431 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
5432 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
5434 free_extent_map(em
);
5438 em
->bdev
= (struct block_device
*)map
;
5439 em
->start
= logical
;
5442 em
->block_start
= 0;
5443 em
->block_len
= em
->len
;
5445 map
->num_stripes
= num_stripes
;
5446 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
5447 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
5448 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
5449 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
5450 map
->type
= btrfs_chunk_type(leaf
, chunk
);
5451 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
5452 for (i
= 0; i
< num_stripes
; i
++) {
5453 map
->stripes
[i
].physical
=
5454 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
5455 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
5456 read_extent_buffer(leaf
, uuid
, (unsigned long)
5457 btrfs_stripe_dev_uuid_nr(chunk
, i
),
5459 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
5461 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
5463 free_extent_map(em
);
5466 if (!map
->stripes
[i
].dev
) {
5467 map
->stripes
[i
].dev
=
5468 add_missing_dev(root
, devid
, uuid
);
5469 if (!map
->stripes
[i
].dev
) {
5471 free_extent_map(em
);
5475 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
5478 write_lock(&map_tree
->map_tree
.lock
);
5479 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
5480 write_unlock(&map_tree
->map_tree
.lock
);
5481 BUG_ON(ret
); /* Tree corruption */
5482 free_extent_map(em
);
5487 static void fill_device_from_item(struct extent_buffer
*leaf
,
5488 struct btrfs_dev_item
*dev_item
,
5489 struct btrfs_device
*device
)
5493 device
->devid
= btrfs_device_id(leaf
, dev_item
);
5494 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
5495 device
->total_bytes
= device
->disk_total_bytes
;
5496 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
5497 device
->type
= btrfs_device_type(leaf
, dev_item
);
5498 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
5499 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
5500 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
5501 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
5502 device
->is_tgtdev_for_dev_replace
= 0;
5504 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
5505 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
5508 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
5510 struct btrfs_fs_devices
*fs_devices
;
5513 BUG_ON(!mutex_is_locked(&uuid_mutex
));
5515 fs_devices
= root
->fs_info
->fs_devices
->seed
;
5516 while (fs_devices
) {
5517 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5521 fs_devices
= fs_devices
->seed
;
5524 fs_devices
= find_fsid(fsid
);
5530 fs_devices
= clone_fs_devices(fs_devices
);
5531 if (IS_ERR(fs_devices
)) {
5532 ret
= PTR_ERR(fs_devices
);
5536 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
5537 root
->fs_info
->bdev_holder
);
5539 free_fs_devices(fs_devices
);
5543 if (!fs_devices
->seeding
) {
5544 __btrfs_close_devices(fs_devices
);
5545 free_fs_devices(fs_devices
);
5550 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
5551 root
->fs_info
->fs_devices
->seed
= fs_devices
;
5556 static int read_one_dev(struct btrfs_root
*root
,
5557 struct extent_buffer
*leaf
,
5558 struct btrfs_dev_item
*dev_item
)
5560 struct btrfs_device
*device
;
5563 u8 fs_uuid
[BTRFS_UUID_SIZE
];
5564 u8 dev_uuid
[BTRFS_UUID_SIZE
];
5566 devid
= btrfs_device_id(leaf
, dev_item
);
5567 read_extent_buffer(leaf
, dev_uuid
,
5568 (unsigned long)btrfs_device_uuid(dev_item
),
5570 read_extent_buffer(leaf
, fs_uuid
,
5571 (unsigned long)btrfs_device_fsid(dev_item
),
5574 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
5575 ret
= open_seed_devices(root
, fs_uuid
);
5576 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
5580 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
5581 if (!device
|| !device
->bdev
) {
5582 if (!btrfs_test_opt(root
, DEGRADED
))
5586 printk(KERN_WARNING
"warning devid %llu missing\n",
5587 (unsigned long long)devid
);
5588 device
= add_missing_dev(root
, devid
, dev_uuid
);
5591 } else if (!device
->missing
) {
5593 * this happens when a device that was properly setup
5594 * in the device info lists suddenly goes bad.
5595 * device->bdev is NULL, and so we have to set
5596 * device->missing to one here
5598 root
->fs_info
->fs_devices
->missing_devices
++;
5599 device
->missing
= 1;
5603 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
5604 BUG_ON(device
->writeable
);
5605 if (device
->generation
!=
5606 btrfs_device_generation(leaf
, dev_item
))
5610 fill_device_from_item(leaf
, dev_item
, device
);
5611 device
->dev_root
= root
->fs_info
->dev_root
;
5612 device
->in_fs_metadata
= 1;
5613 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
5614 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
5615 spin_lock(&root
->fs_info
->free_chunk_lock
);
5616 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
5618 spin_unlock(&root
->fs_info
->free_chunk_lock
);
5624 int btrfs_read_sys_array(struct btrfs_root
*root
)
5626 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
5627 struct extent_buffer
*sb
;
5628 struct btrfs_disk_key
*disk_key
;
5629 struct btrfs_chunk
*chunk
;
5631 unsigned long sb_ptr
;
5637 struct btrfs_key key
;
5639 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
5640 BTRFS_SUPER_INFO_SIZE
);
5643 btrfs_set_buffer_uptodate(sb
);
5644 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
5646 * The sb extent buffer is artifical and just used to read the system array.
5647 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5648 * pages up-to-date when the page is larger: extent does not cover the
5649 * whole page and consequently check_page_uptodate does not find all
5650 * the page's extents up-to-date (the hole beyond sb),
5651 * write_extent_buffer then triggers a WARN_ON.
5653 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5654 * but sb spans only this function. Add an explicit SetPageUptodate call
5655 * to silence the warning eg. on PowerPC 64.
5657 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
5658 SetPageUptodate(sb
->pages
[0]);
5660 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
5661 array_size
= btrfs_super_sys_array_size(super_copy
);
5663 ptr
= super_copy
->sys_chunk_array
;
5664 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
5667 while (cur
< array_size
) {
5668 disk_key
= (struct btrfs_disk_key
*)ptr
;
5669 btrfs_disk_key_to_cpu(&key
, disk_key
);
5671 len
= sizeof(*disk_key
); ptr
+= len
;
5675 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5676 chunk
= (struct btrfs_chunk
*)sb_ptr
;
5677 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
5680 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
5681 len
= btrfs_chunk_item_size(num_stripes
);
5690 free_extent_buffer(sb
);
5694 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
5696 struct btrfs_path
*path
;
5697 struct extent_buffer
*leaf
;
5698 struct btrfs_key key
;
5699 struct btrfs_key found_key
;
5703 root
= root
->fs_info
->chunk_root
;
5705 path
= btrfs_alloc_path();
5709 mutex_lock(&uuid_mutex
);
5712 /* first we search for all of the device items, and then we
5713 * read in all of the chunk items. This way we can create chunk
5714 * mappings that reference all of the devices that are afound
5716 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
5720 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5724 leaf
= path
->nodes
[0];
5725 slot
= path
->slots
[0];
5726 if (slot
>= btrfs_header_nritems(leaf
)) {
5727 ret
= btrfs_next_leaf(root
, path
);
5734 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5735 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5736 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
5738 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
5739 struct btrfs_dev_item
*dev_item
;
5740 dev_item
= btrfs_item_ptr(leaf
, slot
,
5741 struct btrfs_dev_item
);
5742 ret
= read_one_dev(root
, leaf
, dev_item
);
5746 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5747 struct btrfs_chunk
*chunk
;
5748 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
5749 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
5755 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5757 btrfs_release_path(path
);
5762 unlock_chunks(root
);
5763 mutex_unlock(&uuid_mutex
);
5765 btrfs_free_path(path
);
5769 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
5773 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5774 btrfs_dev_stat_reset(dev
, i
);
5777 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
5779 struct btrfs_key key
;
5780 struct btrfs_key found_key
;
5781 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5782 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5783 struct extent_buffer
*eb
;
5786 struct btrfs_device
*device
;
5787 struct btrfs_path
*path
= NULL
;
5790 path
= btrfs_alloc_path();
5796 mutex_lock(&fs_devices
->device_list_mutex
);
5797 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5799 struct btrfs_dev_stats_item
*ptr
;
5802 key
.type
= BTRFS_DEV_STATS_KEY
;
5803 key
.offset
= device
->devid
;
5804 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
5806 __btrfs_reset_dev_stats(device
);
5807 device
->dev_stats_valid
= 1;
5808 btrfs_release_path(path
);
5811 slot
= path
->slots
[0];
5812 eb
= path
->nodes
[0];
5813 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5814 item_size
= btrfs_item_size_nr(eb
, slot
);
5816 ptr
= btrfs_item_ptr(eb
, slot
,
5817 struct btrfs_dev_stats_item
);
5819 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5820 if (item_size
>= (1 + i
) * sizeof(__le64
))
5821 btrfs_dev_stat_set(device
, i
,
5822 btrfs_dev_stats_value(eb
, ptr
, i
));
5824 btrfs_dev_stat_reset(device
, i
);
5827 device
->dev_stats_valid
= 1;
5828 btrfs_dev_stat_print_on_load(device
);
5829 btrfs_release_path(path
);
5831 mutex_unlock(&fs_devices
->device_list_mutex
);
5834 btrfs_free_path(path
);
5835 return ret
< 0 ? ret
: 0;
5838 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
5839 struct btrfs_root
*dev_root
,
5840 struct btrfs_device
*device
)
5842 struct btrfs_path
*path
;
5843 struct btrfs_key key
;
5844 struct extent_buffer
*eb
;
5845 struct btrfs_dev_stats_item
*ptr
;
5850 key
.type
= BTRFS_DEV_STATS_KEY
;
5851 key
.offset
= device
->devid
;
5853 path
= btrfs_alloc_path();
5855 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
5857 printk_in_rcu(KERN_WARNING
"btrfs: error %d while searching for dev_stats item for device %s!\n",
5858 ret
, rcu_str_deref(device
->name
));
5863 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
5864 /* need to delete old one and insert a new one */
5865 ret
= btrfs_del_item(trans
, dev_root
, path
);
5867 printk_in_rcu(KERN_WARNING
"btrfs: delete too small dev_stats item for device %s failed %d!\n",
5868 rcu_str_deref(device
->name
), ret
);
5875 /* need to insert a new item */
5876 btrfs_release_path(path
);
5877 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
5878 &key
, sizeof(*ptr
));
5880 printk_in_rcu(KERN_WARNING
"btrfs: insert dev_stats item for device %s failed %d!\n",
5881 rcu_str_deref(device
->name
), ret
);
5886 eb
= path
->nodes
[0];
5887 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
5888 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5889 btrfs_set_dev_stats_value(eb
, ptr
, i
,
5890 btrfs_dev_stat_read(device
, i
));
5891 btrfs_mark_buffer_dirty(eb
);
5894 btrfs_free_path(path
);
5899 * called from commit_transaction. Writes all changed device stats to disk.
5901 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
5902 struct btrfs_fs_info
*fs_info
)
5904 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5905 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5906 struct btrfs_device
*device
;
5909 mutex_lock(&fs_devices
->device_list_mutex
);
5910 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5911 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
5914 ret
= update_dev_stat_item(trans
, dev_root
, device
);
5916 device
->dev_stats_dirty
= 0;
5918 mutex_unlock(&fs_devices
->device_list_mutex
);
5923 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
5925 btrfs_dev_stat_inc(dev
, index
);
5926 btrfs_dev_stat_print_on_error(dev
);
5929 void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
5931 if (!dev
->dev_stats_valid
)
5933 printk_ratelimited_in_rcu(KERN_ERR
5934 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5935 rcu_str_deref(dev
->name
),
5936 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5937 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5938 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5939 btrfs_dev_stat_read(dev
,
5940 BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5941 btrfs_dev_stat_read(dev
,
5942 BTRFS_DEV_STAT_GENERATION_ERRS
));
5945 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
5949 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5950 if (btrfs_dev_stat_read(dev
, i
) != 0)
5952 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
5953 return; /* all values == 0, suppress message */
5955 printk_in_rcu(KERN_INFO
"btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5956 rcu_str_deref(dev
->name
),
5957 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5958 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5959 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5960 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5961 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
5964 int btrfs_get_dev_stats(struct btrfs_root
*root
,
5965 struct btrfs_ioctl_get_dev_stats
*stats
)
5967 struct btrfs_device
*dev
;
5968 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5971 mutex_lock(&fs_devices
->device_list_mutex
);
5972 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
5973 mutex_unlock(&fs_devices
->device_list_mutex
);
5977 "btrfs: get dev_stats failed, device not found\n");
5979 } else if (!dev
->dev_stats_valid
) {
5981 "btrfs: get dev_stats failed, not yet valid\n");
5983 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
5984 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5985 if (stats
->nr_items
> i
)
5987 btrfs_dev_stat_read_and_reset(dev
, i
);
5989 btrfs_dev_stat_reset(dev
, i
);
5992 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5993 if (stats
->nr_items
> i
)
5994 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
5996 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
5997 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6001 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6003 struct buffer_head
*bh
;
6004 struct btrfs_super_block
*disk_super
;
6006 bh
= btrfs_read_dev_super(device
->bdev
);
6009 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6011 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6012 set_buffer_dirty(bh
);
6013 sync_dirty_buffer(bh
);